参考链接：
https://github.com/Tianxiaomo/pytorch-YOLOv4
模型训练+推理步骤：
1. 下载代码和预训练模型，准备数据
2. 数据预处理：使用data_process的脚本进行自己标注的数据进行处理
3. 训练模型（需要修改train.py/cfg.py/dataset.py/以及tool中关于数据预处理的相关定义，请看代码去改）
4. 测试模型（修改test.py进行预测）
5. pytorch模型转onnx模型（修改pytorch2onnx.py）
6. onnx模型删减（修改dy_resize.py，删减不支持的算子）
7. onnx转om模型（atc命令如下）
8. 测试om模型（修改pyacl代码，本地代码没用等比例缩放，主要修改acl_dvpp.py的数据预处理）
9. 对比结果（把om模型的输出拿出来，放到test_onnx.py / test_om.py中测试，对比本地模型和atlas模型的检测结果）
一、数据处理
数据处理和数据样例请参考上一篇文章：
https://blog.csdn.net/gm_Ergou/article/details/118599118
二、模型训练
这里只贴出修改过的关键代码，其余代码可从参考链接下载：
https://github.com/Tianxiaomo/pytorch-YOLOv4
代码结构：
1.train.py
# -*- coding: utf-8 -*-
import time
import logging
import os, sys, math
import argparse
from collections import deque
import datetime
import copy
import cv2
from tqdm import tqdm
import numpy as np
import torch
import torch.nn as nn
from torch.utils.data import DataLoader
from torch import optim
from torch.nn import functional as F
from tensorboardX import SummaryWriter
from easydict import EasyDict as edict
from dataset import Yolo_dataset
from cfg import Cfg
from models import Yolov4
from tool.darknet2pytorch import Darknet
from tool.tv_reference.utils import collate_fn as val_collate
from tool.tv_reference.coco_utils import convert_to_coco_api
from tool.tv_reference.coco_eval import CocoEvaluator
def bboxes_iou(bboxes_a, bboxes_b, xyxy=True, GIoU=False, DIoU=False, CIoU=False):
if bboxes_a.shape[1] != 4 or bboxes_b.shape[1] != 4:
raise IndexError
if xyxy:
# intersection top left
tl = torch.max(bboxes_a[:, None, :2], bboxes_b[:, :2])
# intersection bottom right
br = torch.min(bboxes_a[:, None, 2:], bboxes_b[:, 2:])
# convex (smallest enclosing box) top left and bottom right
con_tl = torch.min(bboxes_a[:, None, :2], bboxes_b[:, :2])
con_br = torch.max(bboxes_a[:, None, 2:], bboxes_b[:, 2:])
# centerpoint distance squared
rho2 = ((bboxes_a[:, None, 0] + bboxes_a[:, None, 2]) - (bboxes_b[:, 0] + bboxes_b[:, 2])) ** 2 / 4 + (
(bboxes_a[:, None, 1] + bboxes_a[:, None, 3]) - (bboxes_b[:, 1] + bboxes_b[:, 3])) ** 2 / 4
w1 = bboxes_a[:, 2] - bboxes_a[:, 0]
h1 = bboxes_a[:, 3] - bboxes_a[:, 1]
w2 = bboxes_b[:, 2] - bboxes_b[:, 0]
h2 = bboxes_b[:, 3] - bboxes_b[:, 1]
area_a = torch.prod(bboxes_a[:, 2:] - bboxes_a[:, :2], 1)
area_b = torch.prod(bboxes_b[:, 2:] - bboxes_b[:, :2], 1)
else:
# intersection top left
tl = torch.max((bboxes_a[:, None, :2] - bboxes_a[:, None, 2:] / 2),
(bboxes_b[:, :2] - bboxes_b[:, 2:] / 2))
# intersection bottom right
br = torch.min((bboxes_a[:, None, :2] + bboxes_a[:, None, 2:] / 2),
(bboxes_b[:, :2] + bboxes_b[:, 2:] / 2))
# convex (smallest enclosing box) top left and bottom right
con_tl = torch.min((bboxes_a[:, None, :2] - bboxes_a[:, None, 2:] / 2),
(bboxes_b[:, :2] - bboxes_b[:, 2:] / 2))
con_br = torch.max((bboxes_a[:, None, :2] + bboxes_a[:, None, 2:] / 2),
(bboxes_b[:, :2] + bboxes_b[:, 2:] / 2))
# centerpoint distance squared
rho2 = ((bboxes_a[:, None, :2] - bboxes_b[:, :2]) ** 2 / 4).sum(dim=-1)
w1 = bboxes_a[:, 2]
h1 = bboxes_a[:, 3]
w2 = bboxes_b[:, 2]
h2 = bboxes_b[:, 3]
area_a = torch.prod(bboxes_a[:, 2:], 1)
area_b = torch.prod(bboxes_b[:, 2:], 1)
en = (tl < br).type(tl.type()).prod(dim=2)
area_i = torch.prod(br - tl, 2) * en # * ((tl < br).all())
area_u = area_a[:, None] + area_b - area_i
iou = area_i / area_u
if GIoU or DIoU or CIoU:
if GIoU: # Generalized IoU https://arxiv.org/pdf/1902.09630.pdf
area_c = torch.prod(con_br - con_tl, 2) # convex area
return iou - (area_c - area_u) / area_c # GIoU
if DIoU or CIoU: # Distance or Complete IoU https://arxiv.org/abs/1911.08287v1
# convex diagonal squared
c2 = torch.pow(con_br - con_tl, 2).sum(dim=2) + 1e-16
if DIoU:
return iou - rho2 / c2 # DIoU
elif CIoU: # https://github.com/Zzh-tju/DIoU-SSD-pytorch/blob/master/utils/box/box_utils.py#L47
v = (4 / math.pi ** 2) * torch.pow(torch.atan(w1 / h1).unsqueeze(1) - torch.atan(w2 / h2), 2)
with torch.no_grad():
alpha = v / (1 - iou + v)
return iou - (rho2 / c2 + v * alpha) # CIoU
return iou
class Yolo_loss(nn.Module):
def __init__(self, n_classes=18, n_anchors=3, device=None, batch=2):
super(Yolo_loss, self).__init__()
self.device = device
self.strides = [8, 16, 32]
image_size = 416
self.n_classes = n_classes
self.n_anchors = n_anchors
# self.anchors = [[30,92], [45,88], [65,72], [76,150], [121,197], [144,109], [149,314], [279,217], [293,385]]
self.anchors = [[12, 16], [19, 36], [40, 28], [36, 75], [76, 55], [72, 146], [142, 110], [192, 243], [459, 401]] #coco
self.anch_masks = [[0, 1, 2], [3, 4, 5], [6, 7, 8]]
self.ignore_thre = 0.5
self.masked_anchors, self.ref_anchors, self.grid_x, self.grid_y, self.anchor_w, self.anchor_h = [], [], [], [], [], []
for i in range(3):
all_anchors_grid = [(w / self.strides[i], h / self.strides[i]) for w, h in self.anchors]
masked_anchors = np.array([all_anchors_grid[j] for j in self.anch_masks[i]], dtype=np.float32)
ref_anchors = np.zeros((len(all_anchors_grid), 4), dtype=np.float32)
ref_anchors[:, 2:] = np.array(all_anchors_grid, dtype=np.float32)
ref_anchors = torch.from_numpy(ref_anchors)
# calculate pred - xywh obj cls
fsize = image_size // self.strides[i]
grid_x = torch.arange(fsize, dtype=torch.float).repeat(batch, 3, fsize, 1).to(device)
grid_y = torch.arange(fsize, dtype=torch.float).repeat(batch, 3, fsize, 1).permute(0, 1, 3, 2).to(device)
anchor_w = torch.from_numpy(masked_anchors[:, 0]).repeat(batch, fsize, fsize, 1).permute(0, 3, 1, 2).to(
device)
anchor_h = torch.from_numpy(masked_anchors[:, 1]).repeat(batch, fsize, fsize, 1).permute(0, 3, 1, 2).to(
device)
self.masked_anchors.append(masked_anchors)
self.ref_anchors.append(ref_anchors)
self.grid_x.append(grid_x)
self.grid_y.append(grid_y)
self.anchor_w.append(anchor_w)
self.anchor_h.append(anchor_h)
def build_target(self, pred, labels, batchsize, fsize, n_ch, output_id):
# target assignment
tgt_mask = torch.zeros(batchsize, self.n_anchors, fsize, fsize, 4 + self.n_classes).to(device=self.device)
obj_mask = torch.ones(batchsize, self.n_anchors, fsize, fsize).to(device=self.device)
tgt_scale = torch.zeros(batchsize, self.n_anchors, fsize, fsize, 2).to(self.device)
target = torch.zeros(batchsize, self.n_anchors, fsize, fsize, n_ch).to(self.device)
# labels = labels.cpu().data
nlabel = (labels.sum(dim=2) > 0).sum(dim=1) # number of objects
truth_x_all = (labels[:, :, 2] + labels[:, :, 0]) / (self.strides[output_id] * 2)
truth_y_all = (labels[:, :, 3] + labels[:, :, 1]) / (self.strides[output_id] * 2)
truth_w_all = (labels[:, :, 2] - labels[:, :, 0]) / self.strides[output_id]
truth_h_all = (labels[:, :, 3] - labels[:, :, 1]) / self.strides[output_id]
truth_i_all = truth_x_all.to(torch.int16).cpu().numpy()
truth_j_all = truth_y_all.to(torch.int16).cpu().numpy()
for b in range(batchsize):
n = int(nlabel[b])
if n == 0:
continue
truth_box = torch.zeros(n, 4).to(self.device)
truth_box[:n, 2] = truth_w_all[b, :n]
truth_box[:n, 3] = truth_h_all[b, :n]
truth_i = truth_i_all[b, :n]
truth_j = truth_j_all[b, :n]
# calculate iou between truth and reference anchors
anchor_ious_all = bboxes_iou(truth_box.cpu(), self.ref_anchors[output_id], CIoU=True)
# temp = bbox_iou(truth_box.cpu(), self.ref_anchors[output_id])
best_n_all = anchor_ious_all.argmax(dim=1)
best_n = best_n_all % 3
best_n_mask = ((best_n_all == self.anch_masks[output_id][0]) |
(best_n_all == self.anch_masks[output_id][1]) |
(best_n_all == self.anch_masks[output_id][2]))
if sum(best_n_mask) == 0:
continue
truth_box[:n, 0] = truth_x_all[b, :n]
truth_box[:n, 1] = truth_y_all[b, :n]
pred_ious = bboxes_iou(pred[b].view(-1, 4), truth_box, xyxy=False)
pred_best_iou, _ = pred_ious.max(dim=1)
pred_best_iou = (pred_best_iou > self.ignore_thre)
pred_best_iou = pred_best_iou.view(pred[b].shape[:3])
# set mask to zero (ignore) if pred matches truth
obj_mask[b] = ~ pred_best_iou
for ti in range(best_n.shape[0]):
if best_n_mask[ti] == 1:
i, j = truth_i[ti], truth_j[ti]
a = best_n[ti]
obj_mask[b, a, j, i] = 1
tgt_mask[b, a, j, i, :] = 1
target[b, a, j, i, 0] = truth_x_all[b, ti] - truth_x_all[b, ti].to(torch.int16).to(torch.float)
target[b, a, j, i, 1] = truth_y_all[b, ti] - truth_y_all[b, ti].to(torch.int16).to(torch.float)
target[b, a, j, i, 2] = torch.log(
truth_w_all[b, ti] / torch.Tensor(self.masked_anchors[output_id])[best_n[ti], 0] + 1e-16)
target[b, a, j, i, 3] = torch.log(
truth_h_all[b, ti] / torch.Tensor(self.masked_anchors[output_id])[best_n[ti], 1] + 1e-16)
target[b, a, j, i, 4] = 1
target[b, a, j, i, 5 + labels[b, ti, 4].to(torch.int16).cpu().numpy()] = 1
tgt_scale[b, a, j, i, :] = torch.sqrt(2 - truth_w_all[b, ti] * truth_h_all[b, ti] / fsize / fsize)
return obj_mask, tgt_mask, tgt_scale, target
def forward(self, xin, labels=None):
loss, loss_xy, loss_wh, loss_obj, loss_cls, loss_l2 = 0, 0, 0, 0, 0, 0
for output_id, output in enumerate(xin):
batchsize = output.shape[0]
fsize = output.shape[2]
n_ch = 5 + self.n_classes
output = output.view(batchsize, self.n_anchors, n_ch, fsize, fsize)
output = output.permute(0, 1, 3, 4, 2) # .contiguous()
# logistic activation for xy, obj, cls
output[..., np.r_[:2, 4:n_ch]] = torch.sigmoid(output[..., np.r_[:2, 4:n_ch]])
pred = output[..., :4].clone()
pred[..., 0] += self.grid_x[output_id]
pred[..., 1] += self.grid_y[output_id]
pred[..., 2] = torch.exp(pred[..., 2]) * self.anchor_w[output_id].contiguous()
pred[..., 3] = torch.exp(pred[..., 3]) * self.anchor_h[output_id].contiguous()
obj_mask, tgt_mask, tgt_scale, target = self.build_target(pred, labels, batchsize, fsize, n_ch, output_id)
# loss calculation
output[..., 4] *= obj_mask
output[..., np.r_[0:4, 5:n_ch]] *= tgt_mask
output[..., 2:4] *= tgt_scale
target[..., 4] *= obj_mask
target[..., np.r_[0:4, 5:n_ch]] *= tgt_mask
target[..., 2:4] *= tgt_scale
loss_xy += F.binary_cross_entropy(input=output[..., :2], target=target[..., :2],
weight=tgt_scale * tgt_scale, reduction='sum')
loss_wh += F.mse_loss(input=output[..., 2:4], target=target[..., 2:4], reduction='sum') / 2
loss_obj += F.binary_cross_entropy(input=output[..., 4], target=target[..., 4], reduction='sum')
loss_cls += F.binary_cross_entropy(input=output[..., 5:], target=target[..., 5:], reduction='sum')
loss_l2 += F.mse_loss(input=output, target=target, reduction='sum')
loss = loss_xy + loss_wh + loss_obj + loss_cls
return loss, loss_xy, loss_wh, loss_obj, loss_cls, loss_l2
def collate(batch):
images = []
bboxes = []
for img, box in batch:
images.append([img])
bboxes.append([box])
images = np.concatenate(images, axis=0)
images = images.transpose(0, 3, 1, 2)
images = torch.from_numpy(images).div(255.0)
bboxes = np.concatenate(bboxes, axis=0)
bboxes = torch.from_numpy(bboxes)
return images, bboxes
def train(model, device, config, epochs=5, batch_size=1, save_cp=True, log_step=20, img_scale=0.5):
train_dataset = Yolo_dataset(config.train_label, config, train=True)
val_dataset = Yolo_dataset(config.val_label, config, train=False)
n_train = len(train_dataset)
n_val = len(val_dataset)
train_loader = DataLoader(train_dataset, batch_size=config.batch // config.subdivisions, shuffle=True,
num_workers=8, pin_memory=True, drop_last=True, collate_fn=collate)
val_loader = DataLoader(val_dataset, batch_size=config.batch // config.subdivisions, shuffle=True,
num_workers=8, pin_memory=True, drop_last=True, collate_fn=val_collate)
writer = SummaryWriter(log_dir=config.TRAIN_TENSORBOARD_DIR,
filename_suffix=f'OPT_{config.TRAIN_OPTIMIZER}_LR_{config.learning_rate}_BS_{config.batch}_Sub_{config.subdivisions}_Size_{config.width}',
comment=f'OPT_{config.TRAIN_OPTIMIZER}_LR_{config.learning_rate}_BS_{config.batch}_Sub_{config.subdivisions}_Size_{config.width}')
max_itr = config.TRAIN_EPOCHS * n_train
global_step = 0
# learning rate setup
def burnin_schedule(i):
if i < config.burn_in:
factor = pow(i / config.burn_in, 4)
elif i < config.steps[0]:
factor = 1.0
elif i < config.steps[1]:
factor = 0.1
else:
factor = 0.01
return factor
if config.TRAIN_OPTIMIZER.lower() == 'adam':
optimizer = optim.Adam(model.parameters(),lr=config.learning_rate / config.batch,betas=(0.9, 0.999),eps=1e-08)
elif config.TRAIN_OPTIMIZER.lower() == 'sgd':
optimizer = optim.SGD(params=model.parameters(),lr=config.learning_rate / config.batch,momentum=config.momentum,weight_decay=config.decay)
scheduler = optim.lr_scheduler.LambdaLR(optimizer, burnin_schedule)
# scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer, mode='max', verbose=False, patience=6, min_lr=1e-7)
# scheduler = CosineAnnealingWarmRestarts(optimizer, 0.001, 1e-6, 20)
#loss
criterion = Yolo_loss(device=device, batch=config.batch // config.subdivisions, n_classes=config.classes)
# train
saved_models = deque()
model.train()
for epoch in range(epochs+1):
# model.train()
epoch_loss = 0
epoch_step = 0
with tqdm(total=n_train, desc=f'Epoch {epoch + 1}/{epochs}', unit='img', ncols=50) as pbar:
for i, batch in enumerate(train_loader):
global_step += 1
epoch_step += 1
images = batch[0]
bboxes = batch[1]
images = images.to(device=device, dtype=torch.float32)
bboxes = bboxes.to(device=device)
bboxes_pred = model(images)
# loss
loss, loss_xy, loss_wh, loss_obj, loss_cls, loss_l2 = criterion(bboxes_pred, bboxes)
# loss = loss / config.subdivisions
loss.backward()
epoch_loss += loss.item()
if global_step % config.subdivisions == 0:
optimizer.step()
# scheduler.step()
scheduler.step(loss)
model.zero_grad()
# log
if global_step % (log_step * config.subdivisions) == 0:
writer.add_scalar('train/Loss', loss.item(), global_step)
writer.add_scalar('train/loss_xy', loss_xy.item(), global_step)
writer.add_scalar('train/loss_wh', loss_wh.item(), global_step)
writer.add_scalar('train/loss_obj', loss_obj.item(), global_step)
writer.add_scalar('train/loss_cls', loss_cls.item(), global_step)
writer.add_scalar('train/loss_l2', loss_l2.item(), global_step)
# writer.add_scalar('lr', scheduler.get_lr()[0] * config.batch, global_step)
pbar.set_postfix(**{'loss (batch)': loss.item(), 'loss_xy': loss_xy.item(),
'loss_wh': loss_wh.item(),
'loss_obj': loss_obj.item(),
'loss_cls': loss_cls.item(),
'loss_l2': loss_l2.item(),
# 'lr': scheduler.get_lr()[0] * config.batch
})
logging.debug('Train step_{}: loss : {},loss xy : {},loss wh : {},'
'loss obj : {}，loss cls : {},loss l2 : {},lr : {}'
.format(global_step, loss.item(), loss_xy.item(),
loss_wh.item(), loss_obj.item(),
loss_cls.item(), loss_l2.item(),
0.0001
# scheduler.get_lr()[0] * config.batch
))
pbar.update(images.shape[0])
if epoch%10==0:
if cfg.use_darknet_cfg:
eval_model = Darknet(cfg.cfgfile, inference=True)
else:
eval_model = Yolov4(cfg.pretrained, n_classes=cfg.classes, inference=True)
if torch.cuda.device_count() > 1:
eval_model.load_state_dict(model.module.state_dict())
else:
eval_model.load_state_dict(model.state_dict())
eval_model.to(device)
evaluator = evaluate(eval_model, val_loader, config, device)
del eval_model
stats = evaluator.coco_eval['bbox'].stats
writer.add_scalar('train/AP', stats[0], global_step)
writer.add_scalar('train/AP50', stats[1], global_step)
writer.add_scalar('train/AP75', stats[2], global_step)
writer.add_scalar('train/AP_small', stats[3], global_step)
writer.add_scalar('train/AP_medium', stats[4], global_step)
writer.add_scalar('train/AP_large', stats[5], global_step)
writer.add_scalar('train/AR1', stats[6], global_step)
writer.add_scalar('train/AR10', stats[7], global_step)
writer.add_scalar('train/AR100', stats[8], global_step)
writer.add_scalar('train/AR_small', stats[9], global_step)
writer.add_scalar('train/AR_medium', stats[10], global_step)
writer.add_scalar('train/AR_large', stats[11], global_step)
if epoch%20==0:
if not os.path.exists(config.save_path):
os.makedirs(config.save_path)
if not os.path.exists(config.save_path2):
os.makedirs(config.save_path2)
model_name = f'yolov4_{epoch}.pth'
torch.save(model.state_dict(), config.save_path + model_name)
torch.save(model, config.save_path2 + model_name)
logging.info(f'yolov4_{epoch} saved !')
writer.close()
@torch.no_grad()
def evaluate(model, data_loader, cfg, device, logger=None, **kwargs):
""" finished, tested
"""
# cpu_device = torch.device("cpu")
model.eval()
# header = 'Test:'
coco = convert_to_coco_api(data_loader.dataset, bbox_fmt='coco')
coco_evaluator = CocoEvaluator(coco, iou_types = ["bbox"], bbox_fmt='coco')
for images, targets in data_loader:
model_input = [[cv2.resize(img, (cfg.w, cfg.h))] for img in images]
model_input = np.concatenate(model_input, axis=0)
model_input = model_input.transpose(0, 3, 1, 2)
model_input = torch.from_numpy(model_input).div(255.0)
model_input = model_input.to(device)
targets = [{k: v.to(device) for k, v in t.items()} for t in targets]
if torch.cuda.is_available():
torch.cuda.synchronize()
model_time = time.time()
outputs = model(model_input)
# outputs = [{k: v.to(cpu_device) for k, v in t.items()} for t in outputs]
model_time = time.time() - model_time
# outputs = outputs.cpu().detach().numpy()
res = {}
# for img, target, output in zip(images, targets, outputs):
for img, target, boxes, confs in zip(images, targets, outputs[0], outputs[1]):
img_height, img_width = img.shape[:2]
# boxes = output[...,:4].copy() # output boxes in yolo format
boxes = boxes.squeeze(2).cpu().detach().numpy()
boxes[...,2:] = boxes[...,2:] - boxes[...,:2] # Transform [x1, y1, x2, y2] to [x1, y1, w, h]
boxes[...,0] = boxes[...,0]*img_width
boxes[...,1] = boxes[...,1]*img_height
boxes[...,2] = boxes[...,2]*img_width
boxes[...,3] = boxes[...,3]*img_height
boxes = torch.as_tensor(boxes, dtype=torch.float32)
# confs = output[...,4:].copy()
confs = confs.cpu().detach().numpy()
labels = np.argmax(confs, axis=1).flatten()
labels = torch.as_tensor(labels, dtype=torch.int64)
scores = np.max(confs, axis=1).flatten()
scores = torch.as_tensor(scores, dtype=torch.float32)
res[target["image_id"].item()] = {
"boxes": boxes,
"scores": scores,
"labels": labels,
}
evaluator_time = time.time()
coco_evaluator.update(res)
evaluator_time = time.time() - evaluator_time
# gather the stats from all processes
coco_evaluator.synchronize_between_processes()
# accumulate predictions from all images
coco_evaluator.accumulate()
coco_evaluator.summarize()
return coco_evaluator
def init_logger(log_file=None, log_dir=None, log_level=logging.INFO, mode='w', stdout=True):
"""
log_dir: 日志文件的文件夹路径
mode: 'a', append; 'w', 覆盖原文件写入.
"""
def get_date_str():
now = datetime.datetime.now()
return now.strftime('%Y-%m-%d_%H-%M-%S')
fmt = '%(asctime)s %(filename)s[line:%(lineno)d] %(levelname)s: %(message)s'
if log_dir is None:
log_dir = '~/log/'
if log_file is None:
log_file = 'log_' + get_date_str() + '.txt'
if not os.path.exists(log_dir):
os.makedirs(log_dir)
log_file = os.path.join(log_dir, log_file)
# 此处不能使用logging输出
print('log file path:' + log_file)
logging.basicConfig(level=logging.DEBUG,
format=fmt,
filename=log_file,
filemode=mode)
if stdout:
console = logging.StreamHandler(stream=sys.stdout)
console.setLevel(log_level)
formatter = logging.Formatter(fmt)
console.setFormatter(formatter)
logging.getLogger('').addHandler(console)
return logging
def _get_date_str():
now = datetime.datetime.now()
return now.strftime('%Y-%m-%d_%H-%M')
def get_args(**kwargs):
cfg = kwargs
parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument('-l', '--learning-rate', metavar='LR', type=float, nargs='?', default=0.001, dest='learning_rate')
parser.add_argument('-f', '--load', dest='load', type=str, default='data/model/yolov4.pth', help='Load model from a .pth file')
parser.add_argument('-g', '--gpu', metavar='G', type=str, default='-1',help='GPU', dest='gpu')
parser.add_argument('-dir', '--data-dir', type=str, default='', help='dataset dir', dest='dataset_dir')
parser.add_argument('-pretrained', type=str, default='data/model/yolov4.conv.137.pth', help='pretrained yolov4.conv.137')
parser.add_argument('-classes', type=int, default=3, help='dataset classes')
parser.add_argument('-train_label_path', dest='train_label', type=str, default='data/dataset/coins/train.txt')
parser.add_argument('-val_label_path', dest='val_label', type=str, default='data/dataset/coins/train.txt')
parser.add_argument('-optimizer', type=str, default='adam', help='training optimizer', dest='TRAIN_OPTIMIZER')
parser.add_argument('-iou-type', type=str, default='iou', help='iou type (iou, giou, diou, ciou)', dest='iou_type')
parser.add_argument('-TRAIN_EPOCHS', type=int, default=200)
args = vars(parser.parse_args())
# for k in args.keys():
# cfg[k] = args.get(k)
cfg.update(args)
return edict(cfg)
if __name__ == "__main__":
cfg = get_args(**Cfg)
os.environ["CUDA_VISIBLE_DEVICES"] = cfg.gpu
logging = init_logger(log_dir='data/log')
if cfg.use_darknet_cfg:
model = Darknet(cfg.cfgfile)
else:
model = Yolov4(cfg.pretrained, n_classes=cfg.classes)
if torch.cuda.device_count() > 1:
model = torch.nn.DataParallel(model)
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model.to(device=device)
# for param in model.backbone.parameters():
# param.requires_grad = True
train(model=model, config=cfg, epochs=cfg.TRAIN_EPOCHS, device=device)
2.test.py
import torch
from torch import nn
import torch.nn.functional as F
from tool.torch_utils import *
from tool.yolo_layer import YoloLayer
import sys
import cv2
class Mish(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, x):
x = x * (torch.tanh(torch.nn.functional.softplus(x)))
return x
class Upsample(nn.Module):
def __init__(self):
super(Upsample, self).__init__()
def forward(self, x, target_size, inference=False):
assert (x.data.dim() == 4)
# _, _, tH, tW = target_size
if inference:
#B = x.data.size(0)
#C = x.data.size(1)
#H = x.data.size(2)
#W = x.data.size(3)
return x.view(x.size(0), x.size(1), x.size(2), 1, x.size(3), 1).\
expand(x.size(0), x.size(1), x.size(2), target_size[2] // x.size(2), x.size(3), target_size[3] // x.size(3)).\
contiguous().view(x.size(0), x.size(1), target_size[2], target_size[3])
else:
return F.interpolate(x, size=(target_size[2], target_size[3]), mode='nearest')
class Conv_Bn_Activation(nn.Module):
def __init__(self, in_channels, out_channels, kernel_size, stride, activation, bn=True, bias=False):
super().__init__()
pad = (kernel_size - 1) // 2
self.conv = nn.ModuleList()
if bias:
self.conv.append(nn.Conv2d(in_channels, out_channels, kernel_size, stride, pad))
else:
self.conv.append(nn.Conv2d(in_channels, out_channels, kernel_size, stride, pad, bias=False))
if bn:
self.conv.append(nn.BatchNorm2d(out_channels))
if activation == "mish":
self.conv.append(Mish())
elif activation == "relu":
self.conv.append(nn.ReLU(inplace=True))
elif activation == "leaky":
self.conv.append(nn.LeakyReLU(0.1, inplace=True))
elif activation == "linear":
pass
else:
print("activate error !!! {} {} {}".format(sys._getframe().f_code.co_filename,
sys._getframe().f_code.co_name, sys._getframe().f_lineno))
def forward(self, x):
for l in self.conv:
x = l(x)
return x
class ResBlock(nn.Module):
"""
Sequential residual blocks each of which consists of \
two convolution layers.
Args:
ch (int): number of input and output channels.
nblocks (int): number of residual blocks.
shortcut (bool): if True, residual tensor addition is enabled.
"""
def __init__(self, ch, nblocks=1, shortcut=True):
super().__init__()
self.shortcut = shortcut
self.module_list = nn.ModuleList()
for i in range(nblocks):
resblock_one = nn.ModuleList()
resblock_one.append(Conv_Bn_Activation(ch, ch, 1, 1, 'mish'))
resblock_one.append(Conv_Bn_Activation(ch, ch, 3, 1, 'mish'))
self.module_list.append(resblock_one)
def forward(self, x):
for module in self.module_list:
h = x
for res in module:
h = res(h)
x = x + h if self.shortcut else h
return x
class DownSample1(nn.Module):
def __init__(self):
super().__init__()
self.conv1 = Conv_Bn_Activation(3, 32, 3, 1, 'mish')
self.conv2 = Conv_Bn_Activation(32, 64, 3, 2, 'mish')
self.conv3 = Conv_Bn_Activation(64, 64, 1, 1, 'mish')
# [route]
# layers = -2
self.conv4 = Conv_Bn_Activation(64, 64, 1, 1, 'mish')
self.conv5 = Conv_Bn_Activation(64, 32, 1, 1, 'mish')
self.conv6 = Conv_Bn_Activation(32, 64, 3, 1, 'mish')
# [shortcut]
# from=-3
# activation = linear
self.conv7 = Conv_Bn_Activation(64, 64, 1, 1, 'mish')
# [route]
# layers = -1, -7
self.conv8 = Conv_Bn_Activation(128, 64, 1, 1, 'mish')
def forward(self, input):
x1 = self.conv1(input)
x2 = self.conv2(x1)
x3 = self.conv3(x2)
# route -2
x4 = self.conv4(x2)
x5 = self.conv5(x4)
x6 = self.conv6(x5)
# shortcut -3
x6 = x6 + x4
x7 = self.conv7(x6)
# [route]
# layers = -1, -7
x7 = torch.cat([x7, x3], dim=1)
x8 = self.conv8(x7)
return x8
class DownSample2(nn.Module):
def __init__(self):
super().__init__()
self.conv1 = Conv_Bn_Activation(64, 128, 3, 2, 'mish')
self.conv2 = Conv_Bn_Activation(128, 64, 1, 1, 'mish')
# r -2
self.conv3 = Conv_Bn_Activation(128, 64, 1, 1, 'mish')
self.resblock = ResBlock(ch=64, nblocks=2)
# s -3
self.conv4 = Conv_Bn_Activation(64, 64, 1, 1, 'mish')
# r -1 -10
self.conv5 = Conv_Bn_Activation(128, 128, 1, 1, 'mish')
def forward(self, input):
x1 = self.conv1(input)
x2 = self.conv2(x1)
x3 = self.conv3(x1)
r = self.resblock(x3)
x4 = self.conv4(r)
x4 = torch.cat([x4, x2], dim=1)
x5 = self.conv5(x4)
return x5
class DownSample3(nn.Module):
def __init__(self):
super().__init__()
self.conv1 = Conv_Bn_Activation(128, 256, 3, 2, 'mish')
self.conv2 = Conv_Bn_Activation(256, 128, 1, 1, 'mish')
self.conv3 = Conv_Bn_Activation(256, 128, 1, 1, 'mish')
self.resblock = ResBlock(ch=128, nblocks=8)
self.conv4 = Conv_Bn_Activation(128, 128, 1, 1, 'mish')
self.conv5 = Conv_Bn_Activation(256, 256, 1, 1, 'mish')
def forward(self, input):
x1 = self.conv1(input)
x2 = self.conv2(x1)
x3 = self.conv3(x1)
r = self.resblock(x3)
x4 = self.conv4(r)
x4 = torch.cat([x4, x2], dim=1)
x5 = self.conv5(x4)
return x5
class DownSample4(nn.Module):
def __init__(self):
super().__init__()
self.conv1 = Conv_Bn_Activation(256, 512, 3, 2, 'mish')
self.conv2 = Conv_Bn_Activation(512, 256, 1, 1, 'mish')
self.conv3 = Conv_Bn_Activation(512, 256, 1, 1, 'mish')
self.resblock = ResBlock(ch=256, nblocks=8)
self.conv4 = Conv_Bn_Activation(256, 256, 1, 1, 'mish')
self.conv5 = Conv_Bn_Activation(512, 512, 1, 1, 'mish')
def forward(self, input):
x1 = self.conv1(input)
x2 = self.conv2(x1)
x3 = self.conv3(x1)
r = self.resblock(x3)
x4 = self.conv4(r)
x4 = torch.cat([x4, x2], dim=1)
x5 = self.conv5(x4)
return x5
class DownSample5(nn.Module):
def __init__(self):
super().__init__()
self.conv1 = Conv_Bn_Activation(512, 1024, 3, 2, 'mish')
self.conv2 = Conv_Bn_Activation(1024, 512, 1, 1, 'mish')
self.conv3 = Conv_Bn_Activation(1024, 512, 1, 1, 'mish')
self.resblock = ResBlock(ch=512, nblocks=4)
self.conv4 = Conv_Bn_Activation(512, 512, 1, 1, 'mish')
self.conv5 = Conv_Bn_Activation(1024, 1024, 1, 1, 'mish')
def forward(self, input):
x1 = self.conv1(input)
x2 = self.conv2(x1)
x3 = self.conv3(x1)
r = self.resblock(x3)
x4 = self.conv4(r)
x4 = torch.cat([x4, x2], dim=1)
x5 = self.conv5(x4)
return x5
class Neck(nn.Module):
def __init__(self, inference=False):
super().__init__()
self.inference = inference
self.conv1 = Conv_Bn_Activation(1024, 512, 1, 1, 'leaky')
self.conv2 = Conv_Bn_Activation(512, 1024, 3, 1, 'leaky')
self.conv3 = Conv_Bn_Activation(1024, 512, 1, 1, 'leaky')
# SPP
self.maxpool1 = nn.MaxPool2d(kernel_size=5, stride=1, padding=5 // 2)
self.maxpool2 = nn.MaxPool2d(kernel_size=9, stride=1, padding=9 // 2)
self.maxpool3 = nn.MaxPool2d(kernel_size=13, stride=1, padding=13 // 2)
# R -1 -3 -5 -6
# SPP
self.conv4 = Conv_Bn_Activation(2048, 512, 1, 1, 'leaky')
self.conv5 = Conv_Bn_Activation(512, 1024, 3, 1, 'leaky')
self.conv6 = Conv_Bn_Activation(1024, 512, 1, 1, 'leaky')
self.conv7 = Conv_Bn_Activation(512, 256, 1, 1, 'leaky')
# UP
self.upsample1 = Upsample()
# R 85
self.conv8 = Conv_Bn_Activation(512, 256, 1, 1, 'leaky')
# R -1 -3
self.conv9 = Conv_Bn_Activation(512, 256, 1, 1, 'leaky')
self.conv10 = Conv_Bn_Activation(256, 512, 3, 1, 'leaky')
self.conv11 = Conv_Bn_Activation(512, 256, 1, 1, 'leaky')
self.conv12 = Conv_Bn_Activation(256, 512, 3, 1, 'leaky')
self.conv13 = Conv_Bn_Activation(512, 256, 1, 1, 'leaky')
self.conv14 = Conv_Bn_Activation(256, 128, 1, 1, 'leaky')
# UP
self.upsample2 = Upsample()
# R 54
self.conv15 = Conv_Bn_Activation(256, 128, 1, 1, 'leaky')
# R -1 -3
self.conv16 = Conv_Bn_Activation(256, 128, 1, 1, 'leaky')
self.conv17 = Conv_Bn_Activation(128, 256, 3, 1, 'leaky')
self.conv18 = Conv_Bn_Activation(256, 128, 1, 1, 'leaky')
self.conv19 = Conv_Bn_Activation(128, 256, 3, 1, 'leaky')
self.conv20 = Conv_Bn_Activation(256, 128, 1, 1, 'leaky')
def forward(self, input, downsample4, downsample3, inference=False):
x1 = self.conv1(input)
x2 = self.conv2(x1)
x3 = self.conv3(x2)
# SPP
m1 = self.maxpool1(x3)
m2 = self.maxpool2(x3)
m3 = self.maxpool3(x3)
spp = torch.cat([m3, m2, m1, x3], dim=1)
# SPP end
x4 = self.conv4(spp)
x5 = self.conv5(x4)
x6 = self.conv6(x5)
x7 = self.conv7(x6)
# UP
up = self.upsample1(x7, downsample4.size(), self.inference)
# R 85
x8 = self.conv8(downsample4)
# R -1 -3
x8 = torch.cat([x8, up], dim=1)
x9 = self.conv9(x8)
x10 = self.conv10(x9)
x11 = self.conv11(x10)
x12 = self.conv12(x11)
x13 = self.conv13(x12)
x14 = self.conv14(x13)
# UP
up = self.upsample2(x14, downsample3.size(), self.inference)
# R 54
x15 = self.conv15(downsample3)
# R -1 -3
x15 = torch.cat([x15, up], dim=1)
x16 = self.conv16(x15)
x17 = self.conv17(x16)
x18 = self.conv18(x17)
x19 = self.conv19(x18)
x20 = self.conv20(x19)
return x20, x13, x6
class Yolov4Head(nn.Module):
def __init__(self, output_ch, n_classes, inference=False):
super().__init__()
self.inference = inference
self.conv1 = Conv_Bn_Activation(128, 256, 3, 1, 'leaky')
self.conv2 = Conv_Bn_Activation(256, output_ch, 1, 1, 'linear', bn=False, bias=True)
self.yolo1 = YoloLayer(
anchor_mask=[0, 1, 2], num_classes=n_classes,
anchors=[12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401],
num_anchors=9, stride=8)
# R -4
self.conv3 = Conv_Bn_Activation(128, 256, 3, 2, 'leaky')
# R -1 -16
self.conv4 = Conv_Bn_Activation(512, 256, 1, 1, 'leaky')
self.conv5 = Conv_Bn_Activation(256, 512, 3, 1, 'leaky')
self.conv6 = Conv_Bn_Activation(512, 256, 1, 1, 'leaky')
self.conv7 = Conv_Bn_Activation(256, 512, 3, 1, 'leaky')
self.conv8 = Conv_Bn_Activation(512, 256, 1, 1, 'leaky')
self.conv9 = Conv_Bn_Activation(256, 512, 3, 1, 'leaky')
self.conv10 = Conv_Bn_Activation(512, output_ch, 1, 1, 'linear', bn=False, bias=True)
self.yolo2 = YoloLayer(
anchor_mask=[3, 4, 5], num_classes=n_classes,
anchors=[12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401],
num_anchors=9, stride=16)
# R -4
self.conv11 = Conv_Bn_Activation(256, 512, 3, 2, 'leaky')
# R -1 -37
self.conv12 = Conv_Bn_Activation(1024, 512, 1, 1, 'leaky')
self.conv13 = Conv_Bn_Activation(512, 1024, 3, 1, 'leaky')
self.conv14 = Conv_Bn_Activation(1024, 512, 1, 1, 'leaky')
self.conv15 = Conv_Bn_Activation(512, 1024, 3, 1, 'leaky')
self.conv16 = Conv_Bn_Activation(1024, 512, 1, 1, 'leaky')
self.conv17 = Conv_Bn_Activation(512, 1024, 3, 1, 'leaky')
self.conv18 = Conv_Bn_Activation(1024, output_ch, 1, 1, 'linear', bn=False, bias=True)
self.yolo3 = YoloLayer(
anchor_mask=[6, 7, 8], num_classes=n_classes,
anchors=[12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401],
num_anchors=9, stride=32)
def forward(self, input1, input2, input3):
x1 = self.conv1(input1)
x2 = self.conv2(x1)
x3 = self.conv3(input1)
# R -1 -16
x3 = torch.cat([x3, input2], dim=1)
x4 = self.conv4(x3)
x5 = self.conv5(x4)
x6 = self.conv6(x5)
x7 = self.conv7(x6)
x8 = self.conv8(x7)
x9 = self.conv9(x8)
x10 = self.conv10(x9)
# R -4
x11 = self.conv11(x8)
# R -1 -37
x11 = torch.cat([x11, input3], dim=1)
x12 = self.conv12(x11)
x13 = self.conv13(x12)
x14 = self.conv14(x13)
x15 = self.conv15(x14)
x16 = self.conv16(x15)
x17 = self.conv17(x16)
x18 = self.conv18(x17)
if self.inference:
y1 = self.yolo1(x2)
y2 = self.yolo2(x10)
y3 = self.yolo3(x18)
return get_region_boxes([y1, y2, y3])
else:
return [x2, x10, x18]
class Yolov4(nn.Module):
def __init__(self, yolov4conv137weight=None, n_classes=80, inference=False):
super().__init__()
output_ch = (4 + 1 + n_classes) * 3
# backbone
self.down1 = DownSample1()
self.down2 = DownSample2()
self.down3 = DownSample3()
self.down4 = DownSample4()
self.down5 = DownSample5()
# neck
self.neek = Neck(inference)
# yolov4conv137
if yolov4conv137weight:
_model = nn.Sequential(self.down1, self.down2, self.down3, self.down4, self.down5, self.neek)
pretrained_dict = torch.load(yolov4conv137weight)
model_dict = _model.state_dict()
# 1. filter out unnecessary keys
pretrained_dict = {k1: v for (k, v), k1 in zip(pretrained_dict.items(), model_dict)}
# 2. overwrite entries in the existing state dict
model_dict.update(pretrained_dict)
_model.load_state_dict(model_dict)
# head
self.head = Yolov4Head(output_ch, n_classes, inference)
def forward(self, input):
d1 = self.down1(input)
d2 = self.down2(d1)
d3 = self.down3(d2)
d4 = self.down4(d3)
d5 = self.down5(d4)
x20, x13, x6 = self.neek(d5, d4, d3)
output = self.head(x20, x13, x6)
return output
if __name__ == "__main__":
from tool.utils import load_class_names, plot_boxes_cv2
from tool.torch_utils import do_detect
# 参数
width=416
height=416
n_classes=3
imgfile="data/test/test.jpg"
weightfile="data/model1/yolov4_200.pth"
namesfile="data/dataset/coins.names"
# weightfile='data/model/yolov4.pth'
# namesfile="data/dataset/coco.names"
model = Yolov4(yolov4conv137weight=None, n_classes=n_classes, inference=True)
pretrained_dict = torch.load(weightfile, map_location=torch.device('cpu'))
model.load_state_dict(pretrained_dict)
img = cv2.imread(imgfile)
input_img = cv2.resize(img, (width, height))
input_img = cv2.cvtColor(input_img, cv2.COLOR_BGR2RGB)
boxes = do_detect(model, input_img, 0.4, 0.6, use_cuda=False)
if namesfile == None:
if n_classes == 20:
namesfile = 'data/voc.names'
elif n_classes == 80:
namesfile = 'data/coco.names'
else:
print("please give namefile")
class_names = load_class_names(namesfile)
plot_boxes_cv2(img, boxes[0], 'result/predictions.jpg', class_names)
3.cfg.py
# -*- coding: utf-8 -*-
import os
from easydict import EasyDict
Cfg = EasyDict()
Cfg.use_darknet_cfg = False
Cfg.cfgfile = 'cfg/yolov4.cfg'
Cfg.width = 416
Cfg.height = 416
Cfg.classes = 3
Cfg.TRAIN_EPOCHS = 200
Cfg.train_label = 'data/dataset/coins/train.txt'
Cfg.val_label = 'data/dataset/coins/train.txt'
Cfg.TRAIN_OPTIMIZER = 'adam'
Cfg.save_path = 'data/model1/'
Cfg.save_path2 = 'data/model2/'
Cfg.TRAIN_TENSORBOARD_DIR = 'data/log'
Cfg.batch = 1
Cfg.subdivisions = 1
Cfg.channels = 3
Cfg.momentum = 0.949
Cfg.decay = 0.0005
Cfg.angle = 0
Cfg.saturation = 1.5
Cfg.exposure = 1.5
Cfg.hue = .1
Cfg.learning_rate = 0.00261
Cfg.burn_in = 1000
Cfg.max_batches = 500500
Cfg.steps = [400000, 450000]
Cfg.policy = Cfg.steps
Cfg.scales = .1, .1
Cfg.cutmix = 0
Cfg.mosaic = 1
Cfg.letter_box = 0
Cfg.jitter = 0.2
Cfg.track = 0
Cfg.w = Cfg.width
Cfg.h = Cfg.height
Cfg.flip = 1
Cfg.blur = 0
Cfg.gaussian = 0
Cfg.boxes = 60 # box num
Cfg.iou_type = 'iou' # 'giou', 'diou', 'ciou'
if Cfg.mosaic and Cfg.cutmix:
Cfg.mixup = 4
elif Cfg.cutmix:
Cfg.mixup = 2
elif Cfg.mosaic:
Cfg.mixup = 3
Cfg.keep_checkpoint_max = 10
4.dataset.py
# -*- coding: utf-8 -*-
'''
@Time : 2020/05/06 21:09
@Author : Tianxiaomo
@File : dataset.py
@Noice :
@Modificattion :
@Author :
@Time :
@Detail :
'''
import os
import random
import sys
import cv2
import numpy as np
import torch
from torch.utils.data.dataset import Dataset
def rand_uniform_strong(min, max):
if min > max:
swap = min
min = max
max = swap
return random.random() * (max - min) + min
def rand_scale(s):
scale = rand_uniform_strong(1, s)
if random.randint(0, 1) % 2:
return scale
return 1. / scale
def rand_precalc_random(min, max, random_part):
if max < min:
swap = min
min = max
max = swap
return (random_part * (max - min)) + min
def fill_truth_detection(bboxes, num_boxes, classes, flip, dx, dy, sx, sy, net_w, net_h):
if bboxes.shape[0] == 0:
return bboxes, 10000
np.random.shuffle(bboxes)
bboxes[:, 0] -= dx
bboxes[:, 2] -= dx
bboxes[:, 1] -= dy
bboxes[:, 3] -= dy
bboxes[:, 0] = np.clip(bboxes[:, 0], 0, sx)
bboxes[:, 2] = np.clip(bboxes[:, 2], 0, sx)
bboxes[:, 1] = np.clip(bboxes[:, 1], 0, sy)
bboxes[:, 3] = np.clip(bboxes[:, 3], 0, sy)
out_box = list(np.where(((bboxes[:, 1] == sy) & (bboxes[:, 3] == sy)) |
((bboxes[:, 0] == sx) & (bboxes[:, 2] == sx)) |
((bboxes[:, 1] == 0) & (bboxes[:, 3] == 0)) |
((bboxes[:, 0] == 0) & (bboxes[:, 2] == 0)))[0])
list_box = list(range(bboxes.shape[0]))
for i in out_box:
list_box.remove(i)
bboxes = bboxes[list_box]
if bboxes.shape[0] == 0:
return bboxes, 10000
bboxes = bboxes[np.where((bboxes[:, 4] < classes) & (bboxes[:, 4] >= 0))[0]]
if bboxes.shape[0] > num_boxes:
bboxes = bboxes[:num_boxes]
min_w_h = np.array([bboxes[:, 2] - bboxes[:, 0], bboxes[:, 3] - bboxes[:, 1]]).min()
bboxes[:, 0] *= (net_w / sx)
bboxes[:, 2] *= (net_w / sx)
bboxes[:, 1] *= (net_h / sy)
bboxes[:, 3] *= (net_h / sy)
if flip:
temp = net_w - bboxes[:, 0]
bboxes[:, 0] = net_w - bboxes[:, 2]
bboxes[:, 2] = temp
return bboxes, min_w_h
def rect_intersection(a, b):
minx = max(a[0], b[0])
miny = max(a[1], b[1])
maxx = min(a[2], b[2])
maxy = min(a[3], b[3])
return [minx, miny, maxx, maxy]
def image_data_augmentation(mat, w, h, pleft, ptop, swidth, sheight, flip, dhue, dsat, dexp, gaussian_noise, blur, truth):
try:
img = mat
oh, ow, _ = img.shape
pleft, ptop, swidth, sheight = int(pleft), int(ptop), int(swidth), int(sheight)
# crop
src_rect = [pleft, ptop, swidth + pleft, sheight + ptop] # x1,y1,x2,y2
img_rect = [0, 0, ow, oh]
new_src_rect = rect_intersection(src_rect, img_rect) # 交集
dst_rect = [max(0, -pleft), max(0, -ptop), max(0, -pleft) + new_src_rect[2] - new_src_rect[0],
max(0, -ptop) + new_src_rect[3] - new_src_rect[1]]
# cv2.Mat sized
if (src_rect[0] == 0 and src_rect[1] == 0 and src_rect[2] == img.shape[0] and src_rect[3] == img.shape[1]):
sized = cv2.resize(img, (w, h), cv2.INTER_LINEAR)
else:
cropped = np.zeros([sheight, swidth, 3])
cropped[:, :, ] = np.mean(img, axis=(0, 1))
cropped[dst_rect[1]:dst_rect[3], dst_rect[0]:dst_rect[2]] = \
img[new_src_rect[1]:new_src_rect[3], new_src_rect[0]:new_src_rect[2]]
# resize
sized = cv2.resize(cropped, (w, h), cv2.INTER_LINEAR)
# flip
if flip:
# cv2.Mat cropped
sized = cv2.flip(sized, 1) # 0 - x-axis, 1 - y-axis, -1 - both axes (x & y)
# HSV augmentation
# cv2.COLOR_BGR2HSV, cv2.COLOR_RGB2HSV, cv2.COLOR_HSV2BGR, cv2.COLOR_HSV2RGB
if dsat != 1 or dexp != 1 or dhue != 0:
if img.shape[2] >= 3:
hsv_src = cv2.cvtColor(sized.astype(np.float32), cv2.COLOR_RGB2HSV) # RGB to HSV
hsv = cv2.split(hsv_src)
hsv[1] *= dsat
hsv[2] *= dexp
hsv[0] += 179 * dhue
hsv_src = cv2.merge(hsv)
sized = np.clip(cv2.cvtColor(hsv_src, cv2.COLOR_HSV2RGB), 0, 255) # HSV to RGB (the same as previous)
else:
sized *= dexp
if blur:
if blur == 1:
dst = cv2.GaussianBlur(sized, (17, 17), 0)
# cv2.bilateralFilter(sized, dst, 17, 75, 75)
else:
ksize = (blur / 2) * 2 + 1
dst = cv2.GaussianBlur(sized, (ksize, ksize), 0)
if blur == 1:
img_rect = [0, 0, sized.cols, sized.rows]
for b in truth:
left = (b.x - b.w / 2.) * sized.shape[1]
width = b.w * sized.shape[1]
top = (b.y - b.h / 2.) * sized.shape[0]
height = b.h * sized.shape[0]
roi(left, top, width, height)
roi = roi & img_rect
dst[roi[0]:roi[0] + roi[2], roi[1]:roi[1] + roi[3]] = sized[roi[0]:roi[0] + roi[2],
roi[1]:roi[1] + roi[3]]
sized = dst
if gaussian_noise:
noise = np.array(sized.shape)
gaussian_noise = min(gaussian_noise, 127)
gaussian_noise = max(gaussian_noise, 0)
cv2.randn(noise, 0, gaussian_noise) # mean and variance
sized = sized + noise
except:
print("OpenCV can't augment image: " + str(w) + " x " + str(h))
sized = mat
return sized
def filter_truth(bboxes, dx, dy, sx, sy, xd, yd):
bboxes[:, 0] -= dx
bboxes[:, 2] -= dx
bboxes[:, 1] -= dy
bboxes[:, 3] -= dy
bboxes[:, 0] = np.clip(bboxes[:, 0], 0, sx)
bboxes[:, 2] = np.clip(bboxes[:, 2], 0, sx)
bboxes[:, 1] = np.clip(bboxes[:, 1], 0, sy)
bboxes[:, 3] = np.clip(bboxes[:, 3], 0, sy)
out_box = list(np.where(((bboxes[:, 1] == sy) & (bboxes[:, 3] == sy)) |
((bboxes[:, 0] == sx) & (bboxes[:, 2] == sx)) |
((bboxes[:, 1] == 0) & (bboxes[:, 3] == 0)) |
((bboxes[:, 0] == 0) & (bboxes[:, 2] == 0)))[0])
list_box = list(range(bboxes.shape[0]))
for i in out_box:
list_box.remove(i)
bboxes = bboxes[list_box]
bboxes[:, 0] += xd
bboxes[:, 2] += xd
bboxes[:, 1] += yd
bboxes[:, 3] += yd
return bboxes
def blend_truth_mosaic(out_img, img, bboxes, w, h, cut_x, cut_y, i_mixup, left_shift, right_shift, top_shift, bot_shift):
left_shift = min(left_shift, w - cut_x)
top_shift = min(top_shift, h - cut_y)
right_shift = min(right_shift, cut_x)
bot_shift = min(bot_shift, cut_y)
if i_mixup == 0:
bboxes = filter_truth(bboxes, left_shift, top_shift, cut_x, cut_y, 0, 0)
out_img[:cut_y, :cut_x] = img[top_shift:top_shift + cut_y, left_shift:left_shift + cut_x]
if i_mixup == 1:
bboxes = filter_truth(bboxes, cut_x - right_shift, top_shift, w - cut_x, cut_y, cut_x, 0)
out_img[:cut_y, cut_x:] = img[top_shift:top_shift + cut_y, cut_x - right_shift:w - right_shift]
if i_mixup == 2:
bboxes = filter_truth(bboxes, left_shift, cut_y - bot_shift, cut_x, h - cut_y, 0, cut_y)
out_img[cut_y:, :cut_x] = img[cut_y - bot_shift:h - bot_shift, left_shift:left_shift + cut_x]
if i_mixup == 3:
bboxes = filter_truth(bboxes, cut_x - right_shift, cut_y - bot_shift, w - cut_x, h - cut_y, cut_x, cut_y)
out_img[cut_y:, cut_x:] = img[cut_y - bot_shift:h - bot_shift, cut_x - right_shift:w - right_shift]
return out_img, bboxes
def draw_box(img, bboxes):
for b in bboxes:
img = cv2.rectangle(img, (b[0], b[1]), (b[2], b[3]), (0, 255, 0), 2)
return img
class Yolo_dataset(Dataset):
def __init__(self, lable_path, cfg, train=True):
super(Yolo_dataset, self).__init__()
if cfg.mixup == 2:
print("cutmix=1 - isn't supported for Detector")
raise
elif cfg.mixup == 2 and cfg.letter_box:
print("Combination: letter_box=1 & mosaic=1 - isn't supported, use only 1 of these parameters")
raise
self.cfg = cfg
self.train = train
truth = {}
f = open(lable_path, 'r', encoding='utf-8')
for line in f.readlines():
data = line.split(" ")
truth[data[0]] = []
for i in data[1:]:
truth[data[0]].append([int(float(j)) for j in i.split(',')])
self.truth = truth
self.imgs = list(self.truth.keys())
def __len__(self):
return len(self.truth.keys())
def __getitem__(self, index):
if not self.train:
return self._get_val_item(index)
img_path = self.imgs[index]
bboxes = np.array(self.truth.get(img_path), dtype=np.float)
img_path = os.path.join(self.cfg.dataset_dir, img_path)
use_mixup = self.cfg.mixup
if random.randint(0, 1):
use_mixup = 0
if use_mixup == 3:
min_offset = 0.2
cut_x = random.randint(int(self.cfg.w * min_offset), int(self.cfg.w * (1 - min_offset)))
cut_y = random.randint(int(self.cfg.h * min_offset), int(self.cfg.h * (1 - min_offset)))
r1, r2, r3, r4, r_scale = 0, 0, 0, 0, 0
dhue, dsat, dexp, flip, blur = 0, 0, 0, 0, 0
gaussian_noise = 0
out_img = np.zeros([self.cfg.h, self.cfg.w, 3])
out_bboxes = []
for i in range(use_mixup + 1):
if i != 0:
img_path = random.choice(list(self.truth.keys()))
bboxes = np.array(self.truth.get(img_path), dtype=np.float)
img_path = os.path.join(self.cfg.dataset_dir, img_path)
img = cv2.imread(img_path)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
if img is None:
continue
oh, ow, oc = img.shape
dh, dw, dc = np.array(np.array([oh, ow, oc]) * self.cfg.jitter, dtype=np.int)
dhue = rand_uniform_strong(-self.cfg.hue, self.cfg.hue)
dsat = rand_scale(self.cfg.saturation)
dexp = rand_scale(self.cfg.exposure)
pleft = random.randint(-dw, dw)
pright = random.randint(-dw, dw)
ptop = random.randint(-dh, dh)
pbot = random.randint(-dh, dh)
flip = random.randint(0, 1) if self.cfg.flip else 0
if (self.cfg.blur):
tmp_blur = random.randint(0, 2) # 0 - disable, 1 - blur background, 2 - blur the whole image
if tmp_blur == 0:
blur = 0
elif tmp_blur == 1:
blur = 1
else:
blur = self.cfg.blur
if self.cfg.gaussian and random.randint(0, 1):
gaussian_noise = self.cfg.gaussian
else:
gaussian_noise = 0
if self.cfg.letter_box:
img_ar = ow / oh
net_ar = self.cfg.w / self.cfg.h
result_ar = img_ar / net_ar
# print(" ow = %d, oh = %d, w = %d, h = %d, img_ar = %f, net_ar = %f, result_ar = %f \n", ow, oh, w, h, img_ar, net_ar, result_ar);
if result_ar > 1: # sheight - should be increased
oh_tmp = ow / net_ar
delta_h = (oh_tmp - oh) / 2
ptop = ptop - delta_h
pbot = pbot - delta_h
# print(" result_ar = %f, oh_tmp = %f, delta_h = %d, ptop = %f, pbot = %f \n", result_ar, oh_tmp, delta_h, ptop, pbot);
else: # swidth - should be increased
ow_tmp = oh * net_ar
delta_w = (ow_tmp - ow) / 2
pleft = pleft - delta_w
pright = pright - delta_w
# printf(" result_ar = %f, ow_tmp = %f, delta_w = %d, pleft = %f, pright = %f \n", result_ar, ow_tmp, delta_w, pleft, pright);
swidth = ow - pleft - pright
sheight = oh - ptop - pbot
truth, min_w_h = fill_truth_detection(bboxes, self.cfg.boxes, self.cfg.classes, flip, pleft, ptop, swidth,
sheight, self.cfg.w, self.cfg.h)
if (min_w_h / 8) < blur and blur > 1: # disable blur if one of the objects is too small
blur = min_w_h / 8
ai = image_data_augmentation(img, self.cfg.w, self.cfg.h, pleft, ptop, swidth, sheight, flip,
dhue, dsat, dexp, gaussian_noise, blur, truth)
if use_mixup == 0:
out_img = ai
out_bboxes = truth
if use_mixup == 1:
if i == 0:
old_img = ai.copy()
old_truth = truth.copy()
elif i == 1:
out_img = cv2.addWeighted(ai, 0.5, old_img, 0.5)
out_bboxes = np.concatenate([old_truth, truth], axis=0)
elif use_mixup == 3:
if flip:
tmp = pleft
pleft = pright
pright = tmp
left_shift = int(min(cut_x, max(0, (-int(pleft) * self.cfg.w / swidth))))
top_shift = int(min(cut_y, max(0, (-int(ptop) * self.cfg.h / sheight))))
right_shift = int(min((self.cfg.w - cut_x), max(0, (-int(pright) * self.cfg.w / swidth))))
bot_shift = int(min(self.cfg.h - cut_y, max(0, (-int(pbot) * self.cfg.h / sheight))))
out_img, out_bbox = blend_truth_mosaic(out_img, ai, truth.copy(), self.cfg.w, self.cfg.h, cut_x,
cut_y, i, left_shift, right_shift, top_shift, bot_shift)
out_bboxes.append(out_bbox)
if use_mixup == 3:
out_bboxes = np.concatenate(out_bboxes, axis=0)
out_bboxes1 = np.zeros([self.cfg.boxes, 5])
out_bboxes1[:min(out_bboxes.shape[0], self.cfg.boxes)] = out_bboxes[:min(out_bboxes.shape[0], self.cfg.boxes)]
# print(out_bboxes1[:min(out_bboxes.shape[0], self.cfg.boxes)])
return out_img, out_bboxes1
def _get_val_item(self, index):
img_path = self.imgs[index]
bboxes_with_cls_id = np.array(self.truth.get(img_path), dtype=np.float)
img = cv2.imread(os.path.join(self.cfg.dataset_dir, img_path))
# img_height, img_width = img.shape[:2]
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
# img = cv2.resize(img, (self.cfg.w, self.cfg.h))
# img = torch.from_numpy(img.transpose(2, 0, 1)).float().div(255.0).unsqueeze(0)
num_objs = len(bboxes_with_cls_id)
target = {}
# boxes to coco format
boxes = bboxes_with_cls_id[...,:4]
boxes[..., 2:] = boxes[..., 2:] - boxes[..., :2] # box width, box height
target['boxes'] = torch.as_tensor(boxes, dtype=torch.float32)
target['labels'] = torch.as_tensor(bboxes_with_cls_id[...,-1].flatten(), dtype=torch.int64)
target['image_id'] = torch.tensor([get_image_id(img_path)])
target['area'] = (target['boxes'][:,3])*(target['boxes'][:,2])
target['iscrowd'] = torch.zeros((num_objs,), dtype=torch.int64)
return img, target
def get_image_id(filename):
img_id=os.path.splitext(os.path.basename(filename))[0].split('-')[-1]
return int(img_id)
if __name__ == "__main__":
from cfg import Cfg
import matplotlib.pyplot as plt
random.seed(2020)
np.random.seed(2020)
Cfg.dataset_dir = '/mnt/e/Dataset'
dataset = Yolo_dataset(Cfg.train_label, Cfg)
for i in range(100):
out_img, out_bboxes = dataset.__getitem__(i)
a = draw_box(out_img.copy(), out_bboxes.astype(np.int32))
plt.imshow(a.astype(np.int32))
plt.show()
5.models.py
import torch
from torch import nn
import torch.nn.functional as F
from tool.torch_utils import *
from tool.yolo_layer import YoloLayer
import sys
import cv2
class Mish(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, x):
x = x * (torch.tanh(torch.nn.functional.softplus(x)))
return x
class Upsample(nn.Module):
def __init__(self):
super(Upsample, self).__init__()
def forward(self, x, target_size, inference=False):
assert (x.data.dim() == 4)
# _, _, tH, tW = target_size
if inference:
#B = x.data.size(0)
#C = x.data.size(1)
#H = x.data.size(2)
#W = x.data.size(3)
return x.view(x.size(0), x.size(1), x.size(2), 1, x.size(3), 1).\
expand(x.size(0), x.size(1), x.size(2), target_size[2] // x.size(2), x.size(3), target_size[3] // x.size(3)).\
contiguous().view(x.size(0), x.size(1), target_size[2], target_size[3])
else:
return F.interpolate(x, size=(target_size[2], target_size[3]), mode='nearest')
class Conv_Bn_Activation(nn.Module):
def __init__(self, in_channels, out_channels, kernel_size, stride, activation, bn=True, bias=False):
super().__init__()
pad = (kernel_size - 1) // 2
self.conv = nn.ModuleList()
if bias:
self.conv.append(nn.Conv2d(in_channels, out_channels, kernel_size, stride, pad))
else:
self.conv.append(nn.Conv2d(in_channels, out_channels, kernel_size, stride, pad, bias=False))
if bn:
self.conv.append(nn.BatchNorm2d(out_channels))
if activation == "mish":
self.conv.append(Mish())
elif activation == "relu":
self.conv.append(nn.ReLU(inplace=True))
elif activation == "leaky":
self.conv.append(nn.LeakyReLU(0.1, inplace=True))
elif activation == "linear":
pass
else:
print("activate error !!! {} {} {}".format(sys._getframe().f_code.co_filename,
sys._getframe().f_code.co_name, sys._getframe().f_lineno))
def forward(self, x):
for l in self.conv:
x = l(x)
return x
class ResBlock(nn.Module):
"""
Sequential residual blocks each of which consists of \
two convolution layers.
Args:
ch (int): number of input and output channels.
nblocks (int): number of residual blocks.
shortcut (bool): if True, residual tensor addition is enabled.
"""
def __init__(self, ch, nblocks=1, shortcut=True):
super().__init__()
self.shortcut = shortcut
self.module_list = nn.ModuleList()
for i in range(nblocks):
resblock_one = nn.ModuleList()
resblock_one.append(Conv_Bn_Activation(ch, ch, 1, 1, 'mish'))
resblock_one.append(Conv_Bn_Activation(ch, ch, 3, 1, 'mish'))
self.module_list.append(resblock_one)
def forward(self, x):
for module in self.module_list:
h = x
for res in module:
h = res(h)
x = x + h if self.shortcut else h
return x
class DownSample1(nn.Module):
def __init__(self):
super().__init__()
self.conv1 = Conv_Bn_Activation(3, 32, 3, 1, 'mish')
self.conv2 = Conv_Bn_Activation(32, 64, 3, 2, 'mish')
self.conv3 = Conv_Bn_Activation(64, 64, 1, 1, 'mish')
# [route]
# layers = -2
self.conv4 = Conv_Bn_Activation(64, 64, 1, 1, 'mish')
self.conv5 = Conv_Bn_Activation(64, 32, 1, 1, 'mish')
self.conv6 = Conv_Bn_Activation(32, 64, 3, 1, 'mish')
# [shortcut]
# from=-3
# activation = linear
self.conv7 = Conv_Bn_Activation(64, 64, 1, 1, 'mish')
# [route]
# layers = -1, -7
self.conv8 = Conv_Bn_Activation(128, 64, 1, 1, 'mish')
def forward(self, input):
x1 = self.conv1(input)
x2 = self.conv2(x1)
x3 = self.conv3(x2)
# route -2
x4 = self.conv4(x2)
x5 = self.conv5(x4)
x6 = self.conv6(x5)
# shortcut -3
x6 = x6 + x4
x7 = self.conv7(x6)
# [route]
# layers = -1, -7
x7 = torch.cat([x7, x3], dim=1)
x8 = self.conv8(x7)
return x8
class DownSample2(nn.Module):
def __init__(self):
super().__init__()
self.conv1 = Conv_Bn_Activation(64, 128, 3, 2, 'mish')
self.conv2 = Conv_Bn_Activation(128, 64, 1, 1, 'mish')
# r -2
self.conv3 = Conv_Bn_Activation(128, 64, 1, 1, 'mish')
self.resblock = ResBlock(ch=64, nblocks=2)
# s -3
self.conv4 = Conv_Bn_Activation(64, 64, 1, 1, 'mish')
# r -1 -10
self.conv5 = Conv_Bn_Activation(128, 128, 1, 1, 'mish')
def forward(self, input):
x1 = self.conv1(input)
x2 = self.conv2(x1)
x3 = self.conv3(x1)
r = self.resblock(x3)
x4 = self.conv4(r)
x4 = torch.cat([x4, x2], dim=1)
x5 = self.conv5(x4)
return x5
class DownSample3(nn.Module):
def __init__(self):
super().__init__()
self.conv1 = Conv_Bn_Activation(128, 256, 3, 2, 'mish')
self.conv2 = Conv_Bn_Activation(256, 128, 1, 1, 'mish')
self.conv3 = Conv_Bn_Activation(256, 128, 1, 1, 'mish')
self.resblock = ResBlock(ch=128, nblocks=8)
self.conv4 = Conv_Bn_Activation(128, 128, 1, 1, 'mish')
self.conv5 = Conv_Bn_Activation(256, 256, 1, 1, 'mish')
def forward(self, input):
x1 = self.conv1(input)
x2 = self.conv2(x1)
x3 = self.conv3(x1)
r = self.resblock(x3)
x4 = self.conv4(r)
x4 = torch.cat([x4, x2], dim=1)
x5 = self.conv5(x4)
return x5
class DownSample4(nn.Module):
def __init__(self):
super().__init__()
self.conv1 = Conv_Bn_Activation(256, 512, 3, 2, 'mish')
self.conv2 = Conv_Bn_Activation(512, 256, 1, 1, 'mish')
self.conv3 = Conv_Bn_Activation(512, 256, 1, 1, 'mish')
self.resblock = ResBlock(ch=256, nblocks=8)
self.conv4 = Conv_Bn_Activation(256, 256, 1, 1, 'mish')
self.conv5 = Conv_Bn_Activation(512, 512, 1, 1, 'mish')
def forward(self, input):
x1 = self.conv1(input)
x2 = self.conv2(x1)
x3 = self.conv3(x1)
r = self.resblock(x3)
x4 = self.conv4(r)
x4 = torch.cat([x4, x2], dim=1)
x5 = self.conv5(x4)
return x5
class DownSample5(nn.Module):
def __init__(self):
super().__init__()
self.conv1 = Conv_Bn_Activation(512, 1024, 3, 2, 'mish')
self.conv2 = Conv_Bn_Activation(1024, 512, 1, 1, 'mish')
self.conv3 = Conv_Bn_Activation(1024, 512, 1, 1, 'mish')
self.resblock = ResBlock(ch=512, nblocks=4)
self.conv4 = Conv_Bn_Activation(512, 512, 1, 1, 'mish')
self.conv5 = Conv_Bn_Activation(1024, 1024, 1, 1, 'mish')
def forward(self, input):
x1 = self.conv1(input)
x2 = self.conv2(x1)
x3 = self.conv3(x1)
r = self.resblock(x3)
x4 = self.conv4(r)
x4 = torch.cat([x4, x2], dim=1)
x5 = self.conv5(x4)
return x5
class Neck(nn.Module):
def __init__(self, inference=False):
super().__init__()
self.inference = inference
self.conv1 = Conv_Bn_Activation(1024, 512, 1, 1, 'leaky')
self.conv2 = Conv_Bn_Activation(512, 1024, 3, 1, 'leaky')
self.conv3 = Conv_Bn_Activation(1024, 512, 1, 1, 'leaky')
# SPP
self.maxpool1 = nn.MaxPool2d(kernel_size=5, stride=1, padding=5 // 2)
self.maxpool2 = nn.MaxPool2d(kernel_size=9, stride=1, padding=9 // 2)
self.maxpool3 = nn.MaxPool2d(kernel_size=13, stride=1, padding=13 // 2)
# R -1 -3 -5 -6
# SPP
self.conv4 = Conv_Bn_Activation(2048, 512, 1, 1, 'leaky')
self.conv5 = Conv_Bn_Activation(512, 1024, 3, 1, 'leaky')
self.conv6 = Conv_Bn_Activation(1024, 512, 1, 1, 'leaky')
self.conv7 = Conv_Bn_Activation(512, 256, 1, 1, 'leaky')
# UP
self.upsample1 = Upsample()
# R 85
self.conv8 = Conv_Bn_Activation(512, 256, 1, 1, 'leaky')
# R -1 -3
self.conv9 = Conv_Bn_Activation(512, 256, 1, 1, 'leaky')
self.conv10 = Conv_Bn_Activation(256, 512, 3, 1, 'leaky')
self.conv11 = Conv_Bn_Activation(512, 256, 1, 1, 'leaky')
self.conv12 = Conv_Bn_Activation(256, 512, 3, 1, 'leaky')
self.conv13 = Conv_Bn_Activation(512, 256, 1, 1, 'leaky')
self.conv14 = Conv_Bn_Activation(256, 128, 1, 1, 'leaky')
# UP
self.upsample2 = Upsample()
# R 54
self.conv15 = Conv_Bn_Activation(256, 128, 1, 1, 'leaky')
# R -1 -3
self.conv16 = Conv_Bn_Activation(256, 128, 1, 1, 'leaky')
self.conv17 = Conv_Bn_Activation(128, 256, 3, 1, 'leaky')
self.conv18 = Conv_Bn_Activation(256, 128, 1, 1, 'leaky')
self.conv19 = Conv_Bn_Activation(128, 256, 3, 1, 'leaky')
self.conv20 = Conv_Bn_Activation(256, 128, 1, 1, 'leaky')
def forward(self, input, downsample4, downsample3, inference=False):
x1 = self.conv1(input)
x2 = self.conv2(x1)
x3 = self.conv3(x2)
# SPP
m1 = self.maxpool1(x3)
m2 = self.maxpool2(x3)
m3 = self.maxpool3(x3)
spp = torch.cat([m3, m2, m1, x3], dim=1)
# SPP end
x4 = self.conv4(spp)
x5 = self.conv5(x4)
x6 = self.conv6(x5)
x7 = self.conv7(x6)
# UP
up = self.upsample1(x7, downsample4.size(), self.inference)
print(downsample4.size())
print(downsample3.size())
# R 85
x8 = self.conv8(downsample4)
# R -1 -3
x8 = torch.cat([x8, up], dim=1)
x9 = self.conv9(x8)
x10 = self.conv10(x9)
x11 = self.conv11(x10)
x12 = self.conv12(x11)
x13 = self.conv13(x12)
x14 = self.conv14(x13)
# UP
up = self.upsample2(x14, downsample3.size(), self.inference)
# R 54
x15 = self.conv15(downsample3)
# R -1 -3
x15 = torch.cat([x15, up], dim=1)
x16 = self.conv16(x15)
x17 = self.conv17(x16)
x18 = self.conv18(x17)
x19 = self.conv19(x18)
x20 = self.conv20(x19)
return x20, x13, x6
class Yolov4Head(nn.Module):
def __init__(self, output_ch, n_classes, inference=False):
super().__init__()
self.inference = inference
self.conv1 = Conv_Bn_Activation(128, 256, 3, 1, 'leaky')
self.conv2 = Conv_Bn_Activation(256, output_ch, 1, 1, 'linear', bn=False, bias=True)
self.yolo1 = YoloLayer(
anchor_mask=[0, 1, 2], num_classes=n_classes,
anchors=[12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401],
num_anchors=9, stride=8)
# R -4
self.conv3 = Conv_Bn_Activation(128, 256, 3, 2, 'leaky')
# R -1 -16
self.conv4 = Conv_Bn_Activation(512, 256, 1, 1, 'leaky')
self.conv5 = Conv_Bn_Activation(256, 512, 3, 1, 'leaky')
self.conv6 = Conv_Bn_Activation(512, 256, 1, 1, 'leaky')
self.conv7 = Conv_Bn_Activation(256, 512, 3, 1, 'leaky')
self.conv8 = Conv_Bn_Activation(512, 256, 1, 1, 'leaky')
self.conv9 = Conv_Bn_Activation(256, 512, 3, 1, 'leaky')
self.conv10 = Conv_Bn_Activation(512, output_ch, 1, 1, 'linear', bn=False, bias=True)
self.yolo2 = YoloLayer(
anchor_mask=[3, 4, 5], num_classes=n_classes,
anchors=[12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401],
num_anchors=9, stride=16)
# R -4
self.conv11 = Conv_Bn_Activation(256, 512, 3, 2, 'leaky')
# R -1 -37
self.conv12 = Conv_Bn_Activation(1024, 512, 1, 1, 'leaky')
self.conv13 = Conv_Bn_Activation(512, 1024, 3, 1, 'leaky')
self.conv14 = Conv_Bn_Activation(1024, 512, 1, 1, 'leaky')
self.conv15 = Conv_Bn_Activation(512, 1024, 3, 1, 'leaky')
self.conv16 = Conv_Bn_Activation(1024, 512, 1, 1, 'leaky')
self.conv17 = Conv_Bn_Activation(512, 1024, 3, 1, 'leaky')
self.conv18 = Conv_Bn_Activation(1024, output_ch, 1, 1, 'linear', bn=False, bias=True)
self.yolo3 = YoloLayer(
anchor_mask=[6, 7, 8], num_classes=n_classes,
anchors=[12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401],
num_anchors=9, stride=32)
def forward(self, input1, input2, input3):
x1 = self.conv1(input1)
x2 = self.conv2(x1)
x3 = self.conv3(input1)
# R -1 -16
x3 = torch.cat([x3, input2], dim=1)
x4 = self.conv4(x3)
x5 = self.conv5(x4)
x6 = self.conv6(x5)
x7 = self.conv7(x6)
x8 = self.conv8(x7)
x9 = self.conv9(x8)
x10 = self.conv10(x9)
# R -4
x11 = self.conv11(x8)
# R -1 -37
x11 = torch.cat([x11, input3], dim=1)
x12 = self.conv12(x11)
x13 = self.conv13(x12)
x14 = self.conv14(x13)
x15 = self.conv15(x14)
x16 = self.conv16(x15)
x17 = self.conv17(x16)
x18 = self.conv18(x17)
if self.inference:
y1 = self.yolo1(x2)
y2 = self.yolo2(x10)
y3 = self.yolo3(x18)
return get_region_boxes([y1, y2, y3])
else:
return [x2, x10, x18]
class Yolov4(nn.Module):
def __init__(self, yolov4conv137weight=None, n_classes=80, inference=False):
super().__init__()
output_ch = (4 + 1 + n_classes) * 3
# backbone
self.down1 = DownSample1()
self.down2 = DownSample2()
self.down3 = DownSample3()
self.down4 = DownSample4()
self.down5 = DownSample5()
# neck
self.neek = Neck(inference)
# yolov4conv137
if yolov4conv137weight:
_model = nn.Sequential(self.down1, self.down2, self.down3, self.down4, self.down5, self.neek)
pretrained_dict = torch.load(yolov4conv137weight)
model_dict = _model.state_dict()
# 1. filter out unnecessary keys
pretrained_dict = {k1: v for (k, v), k1 in zip(pretrained_dict.items(), model_dict)}
# 2. overwrite entries in the existing state dict
model_dict.update(pretrained_dict)
_model.load_state_dict(model_dict)
# head
self.head = Yolov4Head(output_ch, n_classes, inference)
def forward(self, input):
d1 = self.down1(input)
d2 = self.down2(d1)
d3 = self.down3(d2)
d4 = self.down4(d3)
d5 = self.down5(d4)
x20, x13, x6 = self.neek(d5, d4, d3)
output = self.head(x20, x13, x6)
return output
if __name__ == "__main__":
from tool.utils import load_class_names, plot_boxes_cv2
from tool.torch_utils import do_detect
# 参数
width=608
height=608
n_classes=80
imgfile="data/test/dog.jpg"
# weightfile="data/model1/yolov4_150.pth"
# namesfile="data/dataset/mydata/new_classes.names"
weightfile='data/model/yolov4.pth'
namesfile="data/dataset/coco.names"
model = Yolov4(yolov4conv137weight=None, n_classes=n_classes, inference=True)
pretrained_dict = torch.load(weightfile, map_location=torch.device('cpu'))
model.load_state_dict(pretrained_dict)
img = cv2.imread(imgfile)
input_img = cv2.resize(img, (width, height))
input_img = cv2.cvtColor(input_img, cv2.COLOR_BGR2RGB)
for i in range(2):
boxes = do_detect(model, input_img, 0.4, 0.6, use_cuda=False)
# if namesfile == None:
# if n_classes == 20:
# namesfile = 'data/voc.names'
# elif n_classes == 80:
# namesfile = 'data/coco.names'
# else:
# print("please give namefile")
# class_names = load_class_names(namesfile)
# plot_boxes_cv2(img, boxes[0], 'result/predictions.jpg', class_names)
2271
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
