search

AlibabaResearch / efficientteacher

A Supervised and Semi-Supervised Object Detection Library for YOLO Series
GNU General Public License v3.0
823 stars 149 forks source link

Changing the yolo version from yolov5 to yolov7 facing issues while converting yolov7.pt to efficient.pt #103

Open NirmalVignu opened 1 year ago

NirmalVignu commented 1 year ago

I tried efficient teacher ssod on my custom data it's giving some pretty results, so i decided to test efficient teacher based on yolov7 on my custom data. For that i made the following changes created a map file for yolov7 which is used while executing convert_pt_to_efficient.py to convert yolov7.pt to efficient-yolov7.pt based on yolov7 structure below is the map file content

mapv7.txt

backbone.stage0.cv0 model.0 backbone.stage0.cv1 model.1 backbone.stage0.cv0 model.2 backbone.stage0.cv1 model.3 backbone.elan_0.cv0 model.4 backbone.elan_0.cv1 model.5 backbone.elan_0.cv2 model.6 backbone.elan_0.m0.0 model.7 backbone.elan_0.m0.1 model.8 backbone.elan_0.m1.0 model.9 backbone.concat model.10 backbone.elan_0.m1 model.11 backbone.mp model.12 backbone.elan_1.aug.cv0 model.13 backbone.elan_1.aug.cv1 model.14 backbone.elan_1.aug.cv2 model.15 backbone.concat model.16 backbone.elan_1.cv0 model.17 backbone.elan_1.cv1 model.18 backbone.elan_1.cv2 model.19 backbone.elan_1.m0.0 model.20 backbone.elan_1.m0.1 model.21 backbone.elan_1.m1.0 model.22 backbone.concat model.23 backbone.elan_1.m1.1 model.24 backbone.mp1 model.25 backbone.elan_2.aug.cv0 model.26 backbone.elan_2.aug.cv1 model.27 backbone.elan_2.aug.cv2 model.28 backbone.concat model.29 backbone.elan_2.cv0 model.30 backbone.elan_2.cv1 model.31 backbone.elan_2.cv2 model.32 backbone.elan_2.m0.0 model.33 backbone.elan_2.m0.1 model.34 backbone.elan_2.m1.0 model.35 backbone.concat model.36 backbone.elan_2.m1.1.conv model.37 backbone.mp1 model.38 backbone.elan_3.aug.cv0 model.39 backbone.elan_3.aug.cv1 model.40 backbone.elan_3.aug.cv2 model.41 backbone.concat model.42 backbone.elan_3.cv0 model.43 backbone.elan_3.cv1 model.44 backbone.elan_3.cv2 model.45 backbone.elan_3.m0.0 model.46 backbone.elan_3.m0.1 model.47 backbone.elan_3.m1.0 model.48 backbone.concat model.49 backbone.elan_3.m1.1 model.50 neck.sppcspc model.51 neck.conv1 model.52 neck.upsample1 model.53 neck.conv2 model.54 neck.concat model.55 neck.elan_0.cv0 model.56 neck.elan_0.cv1 model.57 neck.elan_0.cv2 model.58 neck.elan_0.cv3 model.59 neck.elan_0.m0 model.60 neck.elan_0.m1 model.61 neck.concat model.62 neck.elan_0.m2 model.63 neck.conv10 model.64 neck.upsample2 model.65 neck.conv11 model.66 neck.concat model.67 neck.elan_1.cv0 model.68 neck.elan_1.cv1 model.69 neck.elan_1.cv2 model.70 neck.elan_1.cv3 model.71 neck.elan_1.m0 model.72 neck.elan_1.m1 model.73 neck.concat model.74 neck.elan_1.m2 model.75 neck.mp1 model.76 neck.conv19 model.77 neck.conv20 model.78 neck.conv21 model.79 neck.concat model.80 neck.elan_2.cv0 model.81 neck.elan_2.cv1 model.82 neck.elan_2.cv2 model.83 neck.elan_2.cv3 model.84 neck.elan_2.m0 model.85 neck.elan_2.m1 model.86 neck.concat model.87 neck.elan_2.m2 model.88 neck.mp1 model.89 neck.conv29 model.90 neck.conv30 model.91 neck.conv31 model.92 neck.concat model.93 neck.elan_3.cv0 model.94 neck.elan_3.cv1 model.95 neck.elan_3.cv2 model.96 neck.elan_3.cv3 model.97 neck.elan_3.m0 model.98 neck.elan_3.m1 model.99 neck.concat model.100 neck.elan_3.m2 model.101 neck.repconv0 model.102 neck.repconv1 model.103 neck.repconv2 model.104 head model.105

And create a custom yaml file for yolov7

Parameters

project: '/runs_yolov7' adam: False epochs: 300 weights: '' prune_finetune: False linear_lr: True hyp: lr0: 0.01 hsv_h: 0.015 hsv_s: 0.7 hsv_v: 0.4 lrf: 0.1 scale: 0.9 no_aug_epochs: 0 mixup: 0.1 warmup_epochs: 3

Model: depth_multiple: 1.00 # model depth multiple width_multiple: 1.00 # layer channel multiple Backbone: name: 'YoloV7' activation: 'SiLU' Neck: name: 'YoloV7' in_channels: [512, 1024, 1024] out_channels: [256, 512, 1024] activation: 'SiLU' Head: name: 'YoloV7' activation: 'SiLU' anchors: [[10,13, 16,30, 33,23],[30,61, 62,45, 59,119],[116,90, 156,198, 373,326]] # P5/32 Loss: type: 'ComputeLoss' cls: 0.3 obj: 0.7 anchor_t: 4.0

Dataset: data_name: 'coco' train: data/custom_train.txt # 118287 images val: data/custom_val.txt # 5000 images test: data/custom_val.txt # 20288 of 40670 images, submit to https://competitions.codalab.org/competitions/20794^ nc: 2 # number of classes np: 0 #number of keypoints names: [ 'rick', 'morty'] img_size: 640 batch_size: 128

By using the above conf file and map file while i trying to execute convert_efficient_to_yolov5 in convert_pt_to_efficient.py by giving cfg='yolov7_custom.yaml' which is given above and pt_path as 'yolov7.pt' which i stored in another folder and map_path= file path which i given for the above map details. I'm encountering the following Error

Error

**Model summary: 432 layers, 37622682 parameters, 37622682 gradients

load weights from u-yolov5... Traceback (most recent call last): File "/content/efficientteacher/scripts/mula_convertor/convert_pt_to_efficient.py", line 102, in convert_yolov5_to_efficient( '/content/efficientteacher/yolov7.pt', '/content/efficientteacher/configs/sup/public/yolov7l_coco.yaml','efficient-yolov7.pt') File "/content/efficientteacher/scripts/mula_convertor/convert_pt_to_efficient.py", line 43, in convert_yolov5_to_efficient model.load_state_dict(new_yolov5s_weight,strict=False) File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 2041, in load_state_dict raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format( RuntimeError: Error(s) in loading state_dict for Model: size mismatch for backbone.stage0.cv0.conv.weight: copying a param with shape torch.Size([64, 64, 3, 3]) from checkpoint, the shape in current model is torch.Size([32, 3, 3, 3]). size mismatch for backbone.stage0.cv0.bn.weight: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([32]). size mismatch for backbone.stage0.cv0.bn.bias: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([32]). size mismatch for backbone.stage0.cv0.bn.running_mean: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([32]). size mismatch for backbone.stage0.cv0.bn.running_var: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([32]). size mismatch for backbone.stage0.cv1.conv.weight: copying a param with shape torch.Size([128, 64, 3, 3]) from checkpoint, the shape in current model is torch.Size([64, 32, 3, 3]). size mismatch for backbone.stage0.cv1.bn.weight: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([64]). size mismatch for backbone.stage0.cv1.bn.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([64]). size mismatch for backbone.stage0.cv1.bn.running_mean: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([64]). size mismatch for backbone.stage0.cv1.bn.running_var: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([64]). size mismatch for backbone.elan_0.cv2.conv.weight: copying a param with shape torch.Size([64, 64, 3, 3]) from checkpoint, the shape in current model is torch.Size([256, 256, 1, 1]). size mismatch for backbone.elan_0.cv2.bn.weight: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([256]). size mismatch for backbone.elan_0.cv2.bn.bias: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([256]). size mismatch for backbone.elan_0.cv2.bn.running_mean: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([256]). size mismatch for backbone.elan_0.cv2.bn.running_var: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([256]). size mismatch for backbone.elan_1.cv2.conv.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([512, 512, 1, 1]). size mismatch for backbone.elan_1.cv2.bn.weight: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([512]). size mismatch for backbone.elan_1.cv2.bn.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([512]). size mismatch for backbone.elan_1.cv2.bn.running_mean: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([512]). size mismatch for backbone.elan_1.cv2.bn.running_var: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([512]). size mismatch for backbone.elan_1.m1.1.conv.weight: copying a param with shape torch.Size([512, 512, 1, 1]) from checkpoint, the shape in current model is torch.Size([128, 128, 3, 3]). size mismatch for backbone.elan_1.m1.1.bn.weight: copying a param with shape torch.Size([512]) from checkpoint, the shape in current model is torch.Size([128]). size mismatch for backbone.elan_1.m1.1.bn.bias: copying a param with shape torch.Size([512]) from checkpoint, the shape in current model is torch.Size([128]). size mismatch for backbone.elan_1.m1.1.bn.running_mean: copying a param with shape torch.Size([512]) from checkpoint, the shape in current model is torch.Size([128]). size mismatch for backbone.elan_1.m1.1.bn.running_var: copying a param with shape torch.Size([512]) from checkpoint, the shape in current model is torch.Size([128]). size mismatch for backbone.elan_2.cv2.conv.weight: copying a param with shape torch.Size([256, 256, 3, 3]) from checkpoint, the shape in current model is torch.Size([1024, 1024, 1, 1]). size mismatch for backbone.elan_2.cv2.bn.weight: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([1024]). size mismatch for backbone.elan_2.cv2.bn.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([1024]). size mismatch for backbone.elan_2.cv2.bn.running_mean: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([1024]). size mismatch for backbone.elan_2.cv2.bn.running_var: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([1024]). size mismatch for backbone.elan_3.cv2.conv.weight: copying a param with shape torch.Size([256, 256, 3, 3]) from checkpoint, the shape in current model is torch.Size([1024, 1024, 1, 1]). size mismatch for backbone.elan_3.cv2.bn.weight: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([1024]). size mismatch for backbone.elan_3.cv2.bn.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([1024]). size mismatch for backbone.elan_3.cv2.bn.running_mean: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([1024]). size mismatch for backbone.elan_3.cv2.bn.running_var: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([1024]). size mismatch for backbone.elan_3.m1.1.conv.weight: copying a param with shape torch.Size([1024, 1024, 1, 1]) from checkpoint, the shape in current model is torch.Size([256, 256, 3, 3]). size mismatch for backbone.elan_3.m1.1.bn.weight: copying a param with shape torch.Size([1024]) from checkpoint, the shape in current model is torch.Size([256]). size mismatch for backbone.elan_3.m1.1.bn.bias: copying a param with shape torch.Size([1024]) from checkpoint, the shape in current model is torch.Size([256]). size mismatch for backbone.elan_3.m1.1.bn.running_mean: copying a param with shape torch.Size([1024]) from checkpoint, the shape in current model is torch.Size([256]). size mismatch for backbone.elan_3.m1.1.bn.running_var: copying a param with shape torch.Size([1024]) from checkpoint, the shape in current model is torch.Size([256]). size mismatch for neck.elan_0.cv3.conv.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([256, 1024, 1, 1]). size mismatch for neck.elan_0.cv3.bn.weight: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([256]). size mismatch for neck.elan_0.cv3.bn.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([256]). size mismatch for neck.elan_0.cv3.bn.running_mean: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([256]). size mismatch for neck.elan_0.cv3.bn.running_var: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([256]). size mismatch for neck.elan_0.m2.conv.weight: copying a param with shape torch.Size([256, 1024, 1, 1]) from checkpoint, the shape in current model is torch.Size([128, 128, 3, 3]). size mismatch for neck.elan_0.m2.bn.weight: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([128]). size mismatch for neck.elan_0.m2.bn.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([128]). size mismatch for neck.elan_0.m2.bn.running_mean: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([128]). size mismatch for neck.elan_0.m2.bn.running_var: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([128]). size mismatch for neck.elan_1.cv3.conv.weight: copying a param with shape torch.Size([64, 64, 3, 3]) from checkpoint, the shape in current model is torch.Size([128, 512, 1, 1]). size mismatch for neck.elan_1.cv3.bn.weight: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([128]). size mismatch for neck.elan_1.cv3.bn.bias: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([128]). size mismatch for neck.elan_1.cv3.bn.running_mean: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([128]). size mismatch for neck.elan_1.cv3.bn.running_var: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([128]). size mismatch for neck.elan_1.m2.conv.weight: copying a param with shape torch.Size([128, 512, 1, 1]) from checkpoint, the shape in current model is torch.Size([64, 64, 3, 3]). size mismatch for neck.elan_1.m2.bn.weight: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([64]). size mismatch for neck.elan_1.m2.bn.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([64]). size mismatch for neck.elan_1.m2.bn.running_mean: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([64]). size mismatch for neck.elan_1.m2.bn.running_var: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([64]). size mismatch for neck.elan_2.cv3.conv.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([256, 1024, 1, 1]). size mismatch for neck.elan_2.cv3.bn.weight: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([256]). size mismatch for neck.elan_2.cv3.bn.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([256]). size mismatch for neck.elan_2.cv3.bn.running_mean: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([256]). size mismatch for neck.elan_2.cv3.bn.running_var: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([256]). size mismatch for neck.elan_2.m2.conv.weight: copying a param with shape torch.Size([256, 1024, 1, 1]) from checkpoint, the shape in current model is torch.Size([128, 128, 3, 3]). size mismatch for neck.elan_2.m2.bn.weight: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([128]). size mismatch for neck.elan_2.m2.bn.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([128]). size mismatch for neck.elan_2.m2.bn.running_mean: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([128]). size mismatch for neck.elan_2.m2.bn.running_var: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([128]). size mismatch for neck.elan_3.cv3.conv.weight: copying a param with shape torch.Size([256, 256, 3, 3]) from checkpoint, the shape in current model is torch.Size([512, 2048, 1, 1]). size mismatch for neck.elan_3.cv3.bn.weight: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([512]). size mismatch for neck.elan_3.cv3.bn.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([512]). size mismatch for neck.elan_3.cv3.bn.running_mean: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([512]). size mismatch for neck.elan_3.cv3.bn.running_var: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([512]). size mismatch for neck.elan_3.m2.conv.weight: copying a param with shape torch.Size([512, 2048, 1, 1]) from checkpoint, the shape in current model is torch.Size([256, 256, 3, 3]). size mismatch for neck.elan_3.m2.bn.weight: copying a param with shape torch.Size([512]) from checkpoint, the shape in current model is torch.Size([256]). size mismatch for neck.elan_3.m2.bn.bias: copying a param with shape torch.Size([512]) from checkpoint, the shape in current model is torch.Size([256]). size mismatch for neck.elan_3.m2.bn.running_mean: copying a param with shape torch.Size([512]) from checkpoint, the shape in current model is torch.Size([256]). size mismatch for neck.elan_3.m2.bn.running_var: copying a param with shape torch.Size([512]) from checkpoint, the shape in current model is torch.Size([256]).**

Is there any other changes i have to do or any changes done wrong in the above files in order to work with yolov7 based efficient teacher ssod?

wqy19980318 commented 1 year ago

Have you resolved the issue of converting YOLOv7 to efficientteacher format？