作者现有的剪枝，经过测试非常有效，现有对网络修改后，再运行剪枝失败，请指教：# parameters nc: 1 # number of classes depth_multiple: 0.33 # model depth multiple width_multiple: 0.50 # layer channel multiple

anchors

anchors:

• [4,4, 5,5, 5,5] # P2/4
• [5,5, 5,5, 6,6] # P3/8
• [6,6, 7,7, 8,8] # P4/16 #- [116,90, 156,198, 373,326] # P5/16 #- 4,4, 5,5, 5,5, 5,5, 5,5, 6,6, 6,6, 7,7, 8,8

YOLOv5 backbone

backbone:

[from, number, module, args]

[from, number, module, args]

[ [ -1, 1, Conv, [ 64, 6, 2, 2, 1, True, 1.0 ] ], # 0-P1/2 [ -1, 1, Conv, [ 128, 3, 2, None, 1, True, 1.0 ] ], # 1-P2/4 [ -1, 3, C3, [ 128, True, 1, [ 0.5, 0.5 ], [ 1.0, 1.0, 1.0 ], 1.0 ] ], [ -1, 1, Conv, [ 256, 3, 2, None, 1, True, 1.0 ] ], # 3-P3/8 [ -1, 6, C3, [ 256, True, 1, [ 0.5, 0.5 ], [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], 1.0 ] ], [ -1, 1, Conv, [ 512, 3, 2, None, 1, True, 1.0 ] ], # 5-P4/16 [ -1, 9, C3, [ 512, True, 1, [ 0.5, 0.5 ], [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], 1.0 ] ],

[-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 去掉３２倍特征图输出

[ -1, 3, C3, [ 512, True, 1, [ 0.5, 0.5 ], [ 1.0, 1.0, 1.0 ], 1.0 ] ],
[ -1, 1, SPPF, [ 512, 5, 0.5 ] ],  # 9

]

YOLOv5 v6.0 head

head: [[-1, 1, Conv, [512, 1, 1, None, 1, True, 1.0]], [-1, 1, nn.Upsample, [None, 2, 'nearest']], [[-1, 4], 1, Concat, [1]], # cat backbone P3 [-1, 3, C3, [512, False, 1, [0.5, 0.5], [1.0, 1.0, 1.0], 1.0]], # 17 (P3/8-small)

[ -1, 1, Conv, [256, 1, 1, None, 1, True, 1.0]], [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], [ [ -1, 2 ], 1, Concat, [ 1 ] ], # cat backbone P2 160*160 [ -1, 3, C3, [256, False, 1, [0.5, 0.5], [1.0, 1.0, 1.0], 1.0]], # 16 (P2/4-small)

[-1, 1, Conv, [256, 3, 2, None, 1, True, 1.0]], [[-1, 13], 1, Concat, [1]], # cat head P2 [-1, 3, C3, [256, False, 1, [0.5, 0.5], [1.0, 1.0, 1.0], 1.0]], # 20 (P3/16-medium)

[-1, 1, Conv, [256, 3, 2, None, 1, True, 1.0]], [[-1, 9], 1, Concat, [1]], # cat head P4 [-1, 3, C3, [512, False, 1, [0.5, 0.5], [1.0, 1.0, 1.0], 1.0]], # 22 (P4/16-large)

[[16, 19, 22], 1, Detect, [nc, anchors]], # Detect(P2, P3, P4) ]

运行EfficientLite 时出错 错误信息： Traceback (most recent call last): File "E://YOLOv5-Multibackbone-Compression-main/train.py", line 656, in main(opt) File "E:/YOLOv5-Multibackbone-Compression-main/train.py", line 553, in main train(opt.hyp, opt, device, callbacks) File "E:/cYOLOv5-Multibackbone-Compression-main/train.py", line 125, in train model = Model(cfg or ckpt['model'].yaml, ch=3, nc=nc, anchors=hyp.get('anchors')).to(device) # create File "E:\YOLOv5-Multibackbone-Compression-main\models\yolo.py", line 103, in init self.model, self.save = parse_model(deepcopy(self.yaml), ch=[ch], pruning=pruning) # model, savelist File "E:\YOLOv5-Multibackbone-Compression-main\models\yolo.py", line 305, in parse_model m_ = nn.Sequential(*[m(*args) for _ in range(n)]) if n > 1 else m(*args) # module TypeError: init() missing 1 required positional argument: 'drop_connect_rate'

您好，我运行yolov5n，s，m，l都没问题，但是运行yolov5x-TPH，yolov5m-TPH-std，yolov5x时出现错误如下： File "/wangjun/TPH_yolo/train.py", line 656, in main(opt) File "/wangjun/TPH_yolo/train.py", line 553, in main train(opt.hyp, opt, device, callbacks) File "/wangjun/TPH_yolo/train.py", line 132, in train model = Model(cfg, ch=3, nc=nc, anchors=hyp.get('anchors')).to(device) # create File "/wangjun/TPH_yolo/models/yolo.py", line 103, in init self.model, self.save = parse_model(deepcopy(self.yaml), ch=[ch], pruning=pruning) # model, savelist File "/wangjun/TPH_yolo/models/yolo.py", line 305, in parse_model m_ = nn.Sequential(*[m(args) for _ in range(n)]) if n > 1 else m(args) # module File "/wangjun/TPH_yolo/models/common.py", line 419, in init self.m = nn.Sequential([Bottleneck(c1_, c1_, shortcut, g, e=rate[i]) for i in range(n)]) File "/wangjun/TPH_yolo/models/common.py", line 419, in self.m = nn.Sequential([Bottleneck(c1_, c1_, shortcut, g, e=rate[i]) for i in range(n)]) IndexError: list index out of range

File "F:/999yolo/YOLOv5-Multibackbone-Compression-main/models/yolo.py", line 331, in model = Model(opt.cfg).to(device) File "F:/999yolo/YOLOv5-Multibackbone-Compression-main/models/yolo.py", line 112, in init m.stride = torch.tensor([s / x.shape[-2] for x in self.forward(torch.zeros(1, ch, s, s))]) # forward File "F:/999yolo/YOLOv5-Multibackbone-Compression-main/models/yolo.py", line 128, in forward return self._forward_once(x, profile, visualize) # single-scale inference, train File "F:/999yolo/YOLOv5-Multibackbone-Compression-main/models/yolo.py", line 152, in _forward_once x = m(x) # run File "D:\Anaconda3\envs\y55\lib\site-packages\torch\nn\modules\module.py", line 889, in _call_impl result = self.forward(*input, **kwargs) File "F:\999yolo\YOLOv5-Multibackbone-Compression-main\models\common.py", line 142, in forward x_squeezed = nn.AdaptiveAvgPool2d(x, 1) TypeError: init() takes 2 positional arguments but 3 were given

论文中提到 Based on YOLOv5, we only apply transformer encoder blocks in the head part to form Transformer Prediction Head (TPH) and the end of backbone.

但是我看TPH更多是用在了neck层上面，backbones上面增加了一个，而head上面是没有增加TPH。head层上面改变更多的是增加了一个头。

不知道我理解的是否有问题。 谢谢了

您好！

我这边最近尝试复现您开源的基于MQBench下的NNIE量化方式，为了解决在海思芯片上算法量化精度掉点的问题。

然后，我这边复现的时候遇到了如下问题：

1.common.py中C3模块的问题，导致报错，修改为官方v6.0版本的C3模块解决。

2.AttributeError: 'GraphModule' object has no attribute 'module'

暂未解决，我看到其他开发者也遇到了类似问题，如下：

https://github.com/Gumpest/YOLOv5-Multibackbone-Compression/issues/83

能帮忙指导一下吗？谢谢~

请问怎么解决？Traceback (most recent call last): File "pruneEagleEye.py", line 127, in rand_prune_and_eval(model,ignore_idx,opt) File "pruneEagleEye.py", line 59, in rand_prune_and_eval pruned_yaml = update_yaml(pruned_yaml, model, ignore_conv_idx, maskconvdict, opt) File "/hy-nas/yolov5-multibackbone/utils/prune_utils.py", line 263, in update_yaml update_yaml_loop(pruned_yaml,name,maskdict) File "/hy-nas/yolov5-multibackbone/utils/prune_utils.py", line 247, in update_yaml_loop args[-3][1] = 0.5 * maskconvdict[named_m_cv2_conv].sum().item() / (c2*0.5) IndexError: list index out of range

以下是我的训练命令行： nohup python train.py --cfg yolov5xP2CBAM-TPH-BiFPN-SPP.yaml --weights 'weights/yolov5x.pt' --img 640 --data myVisDrone.yaml --hyp data/hyps/hyp.visdrone.yaml --epochs 81 --device 0,1 --batch-size 2 --name '5x_tph_ori_640_526' &

然后我训练了81轮 我的metrics/mAP_0.5:0.95只有0.024101 以下是我的train_batch.jpg

这是我的数据集文件

Train/val/test sets as 1) dir: path/to/imgs, 2) file: path/to/imgs.txt, or 3) list: [path/to/imgs1, path/to/imgs2, ..]

#path: ../datasets/VisDrone # dataset root dir #path: /home/ff/WangZF/datasets/visdrone # dataset root dir train: /home/ff/WangZF/datasets/visdrone/YOLOFormat/train/images # train images (relative to 'path') 6471 images val: /home/ff/WangZF/datasets/visdrone/YOLOFormat/val/images # val images (relative to 'path') 548 images test: /home/ff/WangZF/datasets/visdrone/YOLOFormat/val/images # test images (optional) 1610 images

Classes

nc: 10 # number of classes names: [ 'pedestrian', 'people', 'bicycle', 'car', 'van', 'truck', 'tricycle', 'awning-tricycle', 'bus', 'motor' ] 我不知道是哪里出错了