This repository is forked from https://github.com/eriklindernoren/PyTorch-YOLOv3, I modify it to support multi GPU train.

If you want to use multi GPU train, you should modify these below:

• pad targets in utils/datasets.py -> collate_fn() function line149
• remove pad targets in utils/utils.py -> build_targets() function line 301
• re-calculate batch-index(targets) in utils/utils.py -> build_targets() function line 325
• move model outputs/loss from cpu to gpu in models.py -> Darknet.forward() function line 330
• calculate loss sum from all GPUs in train.py -> line 121
• modify log info in train.py -> line 130

1. mAP计算 论文中是55.3, 原来仓库使用conf_thresh=0.001@IOU=0.5 mAP为53.6

• 修改datasets.py中collate_fn(), 具体原因可查看eriklindernoren/PyTorch-YOLOv3#243 -> mAP为54.9
• 修改utils.py中get_batch_statistics(), 具体原因可查看eriklindernoren/PyTorch-YOLOv3#233 -> mAP为52.6
• 原来仓库作者提供的coco2014 val 5k dataset和coco官网提供的5k val dataset不一致

2. gt box越界错误 fix gt box may cross boundary in utils/utils.py -> build_targets() function line 332

3. 多卡读取weights

４．多卡log信息

• data augmentation
• multi scale train √
• multi GPU train √
• label smooth
• mix up
• cosine lr √
• warmup √
• group normalization (deprecated)

• focal loss
• soft nms (not supported)
• GIOU