RuntimeError: shape mismatch: value tensor of shape [290, 1] cannot be broadcast to indexing result of shape [290]

[IncubatorShokuhou]

loss.py里没有import torch.nn.functional as F.然后这个修完了之后还有问题（数据集是coco128)：

Traceback (most recent call last):
  File "/mnt/data/yolov5_research/train.py", line 740, in <module>
    main(opt)
  File "/mnt/data/yolov5_research/train.py", line 637, in main
    train(opt.hyp, opt, device, callbacks)
  File "/mnt/data/yolov5_research/train.py", line 421, in train
    loss, loss_items = compute_loss_ota(pred, targets.to(device), imgs) if aux_ota_loss else compute_loss(pred, targets.to(device)) # loss scaled by batch_siz
  File "/mnt/data/yolov5_research/utils/loss.py", line 1002, in __call__
    tobj[b, a, gj, gi] = (1.0 - self.gr) + self.gr * iou.detach().clamp(0).type(tobj.dtype)  # iou ratio
RuntimeError: shape mismatch: value tensor of shape [290, 1] cannot be broadcast to indexing result of shape [290]

[positive666]

loss.py里没有import torch.nn.functional as F.然后这个修完了之后还有问题（数据集是coco128)：

Traceback (most recent call last):
  File "/mnt/data/yolov5_research/train.py", line 740, in <module>
    main(opt)
  File "/mnt/data/yolov5_research/train.py", line 637, in main
    train(opt.hyp, opt, device, callbacks)
  File "/mnt/data/yolov5_research/train.py", line 421, in train
    loss, loss_items = compute_loss_ota(pred, targets.to(device), imgs) if aux_ota_loss else compute_loss(pred, targets.to(device)) # loss scaled by batch_siz
  File "/mnt/data/yolov5_research/utils/loss.py", line 1002, in __call__
    tobj[b, a, gj, gi] = (1.0 - self.gr) + self.gr * iou.detach().clamp(0).type(tobj.dtype)  # iou ratio
RuntimeError: shape mismatch: value tensor of shape [290, 1] cannot be broadcast to indexing result of shape [290]

有点仓促这是由于IOU的张量维度问题 不过我完善了一下逻辑可以兼容v5 v7的训练，对于Ota-loss，需要加上ota-match,对于aux-ota，需要加上 aux_ota_loss，嗯 感谢你的反馈 我会尽快修复 另外i这段代码导致P6辅助头训练也是存在问题的 作者刚刚修改也是，计算修复还有个代码的 bug明天看看

[positive666]

我想应该改好了 是新的v5一些写法导致的 和旧版v5基础上集成的v7 有冲突 我测试了可以了应该 现在基本完美兼容了v7 并且v5的代码优化更好 所以代码风格还是以实时版本v5为 准 感谢你的反馈 我也后面会训练验证

[IncubatorShokuhou]

@positive666 谢谢大佬。能跑了。但是还有一些问题：

1. 直接用原版的yolov7-e6e.yaml的话，会出现：

   RuntimeError: Given groups=1, weight of size [80, 3, 3, 3], expected input[1, 12, 128, 128] to have 3 channels, but got 12 channels instead

2. 提示AMP无法使用。我已确定apex安装正常（from apex import amp)
3. 训练时出现:WARNING: NMS time limit 1.060s exceeded =================分割线==================== 问题2的amp问题：更换为最新版本的yolov5的check_amp就ok了

[positive666]

@positive666 谢谢大佬。能跑了。但是还有一些问题：

1. 直接用原版的yolov7-e6e.yaml的话，会出现：

RuntimeError: Given groups=1, weight of size [80, 3, 3, 3], expected input[1, 12, 128, 128] to have 3 channels, but got 12 channels instead

2. 提示AMP无法使用。我已确定apex安装正常（from apex import amp)
3. 训练时出现:WARNING: NMS time limit 1.060s exceeded =================分割线==================== 问题2的amp问题：更换为最新版本的yolov5的check_amp就ok了 q1.你是训练嘛，能否提供下完整的命令 q2.已更新 q3.这个是时间限制：

[IncubatorShokuhou]

@positive666 谢谢大佬。能跑了。但是还有一些问题：

1. 直接用原版的yolov7-e6e.yaml的话，会出现：

RuntimeError: Given groups=1, weight of size [80, 3, 3, 3], expected input[1, 12, 128, 128] to have 3 channels, but got 12 channels instead

2. 提示AMP无法使用。我已确定apex安装正常（from apex import amp)
3. 训练时出现:WARNING: NMS time limit 1.060s exceeded =================分割线==================== 问题2的amp问题：更换为最新版本的yolov5的check_amp就ok了 q1.你是训练嘛，能否提供下完整的命令 q2.已更新 q3.这个是时间限制：

python train.py  --cfg  models/v7_cfg/training/yolov7e6e_原版.yaml --imgsz 640 --weights 'yolov7_training_weights/yolov7-e6e_training.pt'  --data data/我自己的数据集.yaml  --aux_ota_loss  --hyp data/hyps/hyp.scratch-v7-p6.yaml --device 0 --batch-size 16 --epoch 1000 --multi-scale --cos-lr

[positive666]

1.你这个错误我还没复现，用的V7的原始yaml，但是这个项目里我删除了Reorg 2.多尺度训练现在有BUG

[IncubatorShokuhou]

Reorg

我把Reorg重新加进去common.py里了。然后就报了这个错。不加的话直接就没法运行，提示缺少Reorg。

[positive666]

Reorg

我把Reorg重新加进去common.py里了。然后就报了这个错。不加的话直接就没法运行，提示缺少Reorg。 yolo.py elif m is ReOrg: c2 = ch[f] * 4

[positive666]

在yolo.py中加上 输出通道数的设置就可以了

[github-actions[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[Maple1024]

loss.py里没有import torch.nn.functional as F.然后这个修完了之后还有问题（数据集是coco128)：

Traceback (most recent call last):
  File "/mnt/data/yolov5_research/train.py", line 740, in <module>
    main(opt)
  File "/mnt/data/yolov5_research/train.py", line 637, in main
    train(opt.hyp, opt, device, callbacks)
  File "/mnt/data/yolov5_research/train.py", line 421, in train
    loss, loss_items = compute_loss_ota(pred, targets.to(device), imgs) if aux_ota_loss else compute_loss(pred, targets.to(device)) # loss scaled by batch_siz
  File "/mnt/data/yolov5_research/utils/loss.py", line 1002, in __call__
    tobj[b, a, gj, gi] = (1.0 - self.gr) + self.gr * iou.detach().clamp(0).type(tobj.dtype)  # iou ratio
RuntimeError: shape mismatch: value tensor of shape [290, 1] cannot be broadcast to indexing result of shape [290]

这个问题是修改了代码哪个地方啊