跑通pytorch版YOLO v3并训练自己的数据

一、环境配置

系统配置——ubuntu16.04
显卡——RTX2080Ti
cuda10.0+cudnn7.5.0
Python3.6+Anaconda
PyTorch1.0.0

二、相关准备

clone代码

git clone https://github.com/eriklindernoren/PyTorch-YOLOv3

安装相关包

pip install -r requirements.txt

下载预训练权重

cd weights
bash download_weights.sh
cd ../

下载coco数据集

cd data
bash get_coco_dataset.sh
cd ../

train2014.zip和val2014.zip的下载太慢，镜像网站https://pjreddie.com/projects/coco-mirror/最好是迅雷下载
用训练的模型test

python3 test.py --weights_path weights/yolov3.weights

AP=0.5145

训练

$ train.py [-h] [--epochs EPOCHS] [--batch_size BATCH_SIZE]
                [--gradient_accumulations GRADIENT_ACCUMULATIONS]
                [--model_def MODEL_DEF] [--data_config DATA_CONFIG]
                [--pretrained_weights PRETRAINED_WEIGHTS] [--n_cpu N_CPU]
                [--img_size IMG_SIZE]
                [--checkpoint_interval CHECKPOINT_INTERVAL]
                [--evaluation_interval EVALUATION_INTERVAL]
                [--compute_map COMPUTE_MAP]
                [--multiscale_training MULTISCALE_TRAINING]

To train on COCO using a Darknet-53 backend pretrained on ImageNet run:

python3 train.py --data_config config/coco.data  --pretrained_weights weights/darknet53.conv.74

运行过程中遇到一个错误

OSError: image file is truncated (28 bytes not processed)

解决方法

from PIL import ImageFile
ImageFile.LOAD_TRUNCATED_IMAGES = True

可视化Tensorboard

tensorboard --logdir='logs' --port=6006

有错误

ValueError: Not a TBLoader or TBPlugin subclass: <class 'tensorboard_plugin_wit.wit_plugin_loader.WhatIfToolPluginLoader'>

解决

pip uninstall tensorboard-plugin-wit

待更新……