yolact参数量和计算量FLOPs统计

一、使用的代码

import torch
from thop import profile
def model_info(model, verbose=False):
    #打印模型参数、计算量
    #train阶段会输出迭代数，等于param
    n_p = sum(x.numel() for x in model.parameters())  # number parameters
    n_g = sum(x.numel() for x in model.parameters() if x.requires_grad)  # number gradients
    if verbose:
        print(f"{'layer':>5} {'name':>40} {'gradient':>9} {'parameters':>12} {'shape':>20} {'mu':>10} {'sigma':>10}")
        for i, (name, p) in enumerate(model.named_parameters()):
            name = name.replace('module_list.', '')
            print('%5g %40s %9s %12g %20s %10.3g %10.3g' %
                  (i, name, p.requires_grad, p.numel(), list(p.shape), p.mean(), p.std()))
    name = 'yolact'
    print(f"{name} summary: {len(list(model.modules()))} layers, {n_p} parameters, {n_g} gradients")
    
def summary_params_flpos():
    net = Yolact()
    net.load_weights('weights/yolact_base_54_800000.pth')
    net.cuda().eval()
    model_info(net)
    img = cv2.imread('yolact-master/data/img.jpg')
    img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
    img = cv2.resize(img, (550, 550))
    img = torch.from_numpy(img).permute(2, 0, 1).unsqueeze(0).float().cuda()
    flops, params = thop.profile(net, inputs=(img, ))

    print('flops: ', flops, 'params: ', params)
    print('flops: %.2f M, params: %.2f M' % (flops / 1000000.0, params / 1000000.0))
summary_params_flpos()

二、问题记录

1. RuntimeError: Can’t add a new parameter after ScriptModule construction. Tried to add 'total_ops

问题描述：运行时错误：不能在ScriptModule构造后添加新参数。试图添加total_ops。
问题分析：由于我之前阅读过代码，知道ScriptModule在模型构造阶段用到了。

class FPN(ScriptModuleWrapper):

接着点进去看 yolact.py 里ScriptModuleWrapper的定义：

# As of March 10, 2019, Pytorch DataParallel still doesn't support JIT Script Modules
use_jit = torch.cuda.device_count() <= 1
if not use_jit:
    print('Multiple GPUs detected! Turning off JIT.')

ScriptModuleWrapper = torch.jit.ScriptModule if use_jit else nn.Module
script_method_wrapper = torch.jit.script_method if use_jit else lambda fn, _rcn=None: fn

模型use_jit参数设置为了True。不知道是什么东东，但是用到了JIT，感觉是trance一些东西，反正和nn.Module 不同。那么设置为False看看，问题解决了。

问题解决：将 torch.jit.ScriptModule 模型修改为 nn.Module。具体设置：在yolact.py中，以下位置添加一行代码 use_jit = False：

use_jit = torch.cuda.device_count() <= 1
use_jit = False  #在这里添加这行
if not use_jit:
    print('Multiple GPUs detected! Turning off JIT.')

ScriptModuleWrapper = torch.jit.ScriptModule if use_jit else nn.Module
script_method_wrapper = torch.jit.script_method if use_jit else lambda fn, _rcn=None: fn

2. RuntimeError: legacy constructor expects device type: cpu but device type: cuda was passed

问题解决：参考https://blog.csdn.net/dong_liuqi/article/details/120099335
用tensor创建而不是Tensor：

#self.priors = torch.Tensor(prior_data, device=device).view(-1, 4).detach()
self.priors = torch.tensor(prior_data, device=device).view(-1, 4).detach()

三、参数量和浮点数统计结果

yolact backbone=resnet101的FLOPs为 82.74GB，参数量为50.14MB。