How to Replace backbone

[DragonBoyL]

I want to change backbone to ResNet, how should I modify the code

[KRK11]

I want to change backbone to ResNet, how should I modify the code

Modify backbone.py to return four feature layers. Of course, only the last three feature layers are used.

[jone222]

File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 2152, in load_state_dict raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format( RuntimeError: Error(s) in loading state_dict for VGG: Missing key(s) in state_dict: "features.1.weight", "features.1.bias", "features.1.running_mean", "features.1.running_var", "features.3.weight", "features.3.bias", "features.4.weight", "features.4.bias", "features.4.running_mean", "features.4.running_var", "features.8.weight", "features.8.bias", "features.8.running_mean", "features.8.running_var", "features.11.weight", "features.11.bias", "features.11.running_mean", "features.11.running_var", "features.15.weight", "features.15.bias", "features.15.running_mean", "features.15.running_var", "features.18.weight", "features.18.bias", "features.18.running_mean", "features.18.running_var", "features.20.weight", "features.20.bias", "features.21.running_mean", "features.21.running_var", "features.25.weight", "features.25.bias", "features.25.running_mean", "features.25.running_var", "features.27.weight", "features.27.bias", "features.28.running_mean", "features.28.running_var", "features.30.weight", "features.30.bias", "features.31.weight", "features.31.bias", "features.31.running_mean", "features.31.running_var", "features.34.weight", "features.34.bias", "features.35.weight", "features.35.bias", "features.35.running_mean", "features.35.running_var", "features.37.weight", "features.37.bias", "features.38.weight", "features.38.bias", "features.38.running_mean", "features.38.running_var", "features.40.weight", "features.40.bias", "features.41.weight", "features.41.bias", "features.41.running_mean", "features.41.running_var". Unexpected key(s) in state_dict: "features.2.weight", "features.2.bias", "features.5.weight", "features.5.bias", "features.12.weight", "features.12.bias", "features.19.weight", "features.19.bias", "features.26.weight", "features.26.bias". size mismatch for features.7.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([128, 64, 3, 3]). size mismatch for features.10.weight: copying a param with shape torch.Size([256, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([128, 128, 3, 3]). size mismatch for features.10.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([128]). size mismatch for features.14.weight: copying a param with shape torch.Size([256, 256, 3, 3]) from checkpoint, the shape in current model is torch.Size([256, 128, 3, 3]). size mismatch for features.17.weight: copying a param with shape torch.Size([512, 256, 3, 3]) from checkpoint, the shape in current model is torch.Size([256, 256, 3, 3]). size mismatch for features.17.bias: copying a param with shape torch.Size([512]) from checkpoint, the shape in current model is torch.Size([256]). size mismatch for features.21.weight: copying a param with shape torch.Size([512, 512, 3, 3]) from checkpoint, the shape in current model is torch.Size([256]). size mismatch for features.21.bias: copying a param with shape torch.Size([512]) from checkpoint, the shape in current model is torch.Size([256]). size mismatch for features.24.weight: copying a param with shape torch.Size([512, 512, 3, 3]) from checkpoint, the shape in current model is torch.Size([512, 256, 3, 3]). size mismatch for features.28.weight: copying a param with shape torch.Size([512, 512, 3, 3]) from checkpoint, the shape in current model is torch.Size([512]).

这个问题怎么解决呀

[KRK11]

你是换了backbone之后直接load之前原模型训练出来的checkpoint吗

[jone222]

没有佬，我刚开始接触，我想问一下，这个模型用于麦穗计数需要改的代码多么

[KRK11]

你甚至可以不用改动，把数据集处理好就可以。

[jone222]

好的佬，我估计数据集没处理好，我再看看。谢谢佬

[jone222]

佬，可以方便加个v吗，后续有些问题的话也想问你，万分感谢，我的v，yjh1203-

[KRK11]

在这问吧，这样别人也可以看到你的问题。

[jone222]

好的，真的是万分感谢

[jone222]

Traceback (most recent call last): File "/content/drive/MyDrive/CrowdCounting-P2PNet-main/train.py", line 222, in main(args) File "/content/drive/MyDrive/CrowdCounting-P2PNet-main/train.py", line 159, in main stat = train_one_epoch( File "/content/drive/MyDrive/CrowdCounting-P2PNet-main/engine.py", line 85, in train_one_epoch for samples, targets in data_loader: File "/usr/local/lib/python3.10/dist-packages/torch/utils/data/dataloader.py", line 630, in next data = self._next_data() File "/usr/local/lib/python3.10/dist-packages/torch/utils/data/dataloader.py", line 1345, in _next_data return self._process_data(data) File "/usr/local/lib/python3.10/dist-packages/torch/utils/data/dataloader.py", line 1371, in _process_data data.reraise() File "/usr/local/lib/python3.10/dist-packages/torch/_utils.py", line 694, in reraise raise exception FileNotFoundError: Caught FileNotFoundError in DataLoader worker process 0.

请问这个错误如何解决，已经导入了数据集，并且更换了train和shha上的数据集路径

[KRK11]

你是不是train.list和test.list没写对，或者是SHHA.py下的init里的self.train_list,self.test_list没对应好

[jone222]

我直接用的小麦的数据集，这张照片是我的数据集格式 在shha.py里面他的原码是self.val=一个测试集的list。需要把这个val改为test吗

[KRK11]

self.train_lists = "train.list" self.eval_list = "test.list" 这样就行，你的数据集train.list里怎么写的，然后gt文件里每行一对点坐标。

[jone222]

train.list是右边部分 我的gt文件是图片对应的txt文件 但是在test文件夹中，我没有划分image和txt直接把他们放在一起了

[KRK11]

gt文件错了，是test/gt/XXXX.txt，不是"text"/gt/XXXX.txt。而且关于gt路径要给对，你test文件夹里得有这个gt文件路径，只要能指定正确路径就行。

[jone222]

好的佬，我再改一下，感谢感谢

[jone222]

我将text改成test之后依旧是这个错误，我想问一下下面图片中的default应该填什么呀

[KRK11]

默认'SHHA'就行，因为源码里的dataload部分实现的时候只考虑了SHHA，只有默认这个才会导入数据。或者你将crowd_datasets/init.py中的if args.dataset_file == 'SHHA':直接删掉，或者改成任意你想要的。

[jone222]

好的，我用SHHA，但是还之前的问题 Traceback (most recent call last): File "/content/drive/MyDrive/CrowdCounting-P2PNet-main/train.py", line 222, in main(args) File "/content/drive/MyDrive/CrowdCounting-P2PNet-main/train.py", line 159, in main stat = train_one_epoch( File "/content/drive/MyDrive/CrowdCounting-P2PNet-main/engine.py", line 85, in train_one_epoch for samples, targets in data_loader: File "/usr/local/lib/python3.10/dist-packages/torch/utils/data/dataloader.py", line 630, in next data = self._next_data() File "/usr/local/lib/python3.10/dist-packages/torch/utils/data/dataloader.py", line 1345, in _next_data return self._process_data(data) File "/usr/local/lib/python3.10/dist-packages/torch/utils/data/dataloader.py", line 1371, in _process_data data.reraise() File "/usr/local/lib/python3.10/dist-packages/torch/_utils.py", line 694, in reraise raise exception FileNotFoundError: Caught FileNotFoundError in DataLoader worker process 0. Original Traceback (most recent call last): File "/usr/local/lib/python3.10/dist-packages/torch/utils/data/_utils/worker.py", line 308, in _worker_loop data = fetcher.fetch(index) File "/usr/local/lib/python3.10/dist-packages/torch/utils/data/_utils/fetch.py", line 51, in fetch data = [self.dataset[idx] for idx in possibly_batched_index] File "/usr/local/lib/python3.10/dist-packages/torch/utils/data/_utils/fetch.py", line 51, in data = [self.dataset[idx] for idx in possibly_batched_index] File "/content/drive/MyDrive/CrowdCounting-P2PNet-main/crowd_datasets/SHHA/SHHA.py", line 53, in getitem img, point = load_data((img_path, gt_path), self.train) File "/content/drive/MyDrive/CrowdCounting-P2PNet-main/crowd_datasets/SHHA/SHHA.py", line 102, in load_data with open(gt_path) as f_label: FileNotFoundError: [Errno 2] No such file or directory: '/content/drive/MyDrive/CrowdCounting-P2PNet-main/DATAROOT/test/gt/0078.txt'

[jone222]

我是不是要根据这个把我数据集中文件名“gt”改为img_gt_path

[KRK11]

不是，是将文件中gt文件对应的真实地址写对，也就是相对地址要对。

[KRK11]

'/content/drive/MyDrive/CrowdCounting-P2PNet-main/DATAROOT/test/gt/0078.txt‘你确定有这个地址吗

[jone222]

这个地址在train中有，是/content/drive/MyDrive/CrowdCounting-P2PNet-main/DATAROOT/train/gt/0078.txt，有这个地址。 但是不明白他为啥在说在test找不到，test没有这个地址，我的train是1-700的图片，test是5134-5833（数字是图片的文件名），在test中没有78

[KRK11]

没有的话你的数据集test.list这个文件写错了，这个文件要的是你测试集里面文件的地址。

[jone222]

我的test.list里面也没有78这个，里面的东西和图片是对应着的呀

[KRK11]

你的train.list里可能出现了这个地址，你是训练还没开始就报错吗，还是在验证时报错

[KRK11]

你的train.list把test后面的gt去掉，因为你直接平铺的，没有gt文件夹

[jone222]

是的，我发现问题是在train.list中，gt的路径那个应该是train下面，不是test下面，谢谢佬，耽误你时间了。现在好像跑起来了 Averaged stats: lr: 0.000100 loss: 0.0112 (0.0842) loss_ce: 0.0112 (0.0842) loss_ce_unscaled: 0.0112 (0.0842) loss_point_unscaled: 3.7103 (4.0512) [ep 0][lr 0.0001000][88.10s] Averaged stats: lr: 0.000100 loss: 0.0084 (0.0093) loss_ce: 0.0084 (0.0093) loss_ce_unscaled: 0.0084 (0.0093) loss_point_unscaled: 3.5111 (3.5377) [ep 1][lr 0.0001000][63.98s 是这样吗，那mse，在哪看呀

[KRK11]

每5轮输出结果。

[jone222]

佬，还在吗，请问一下跑上海或者nwpu这些数据集时候，数据中的mat文件怎么转成txt

[KRK11]

from scipy.io import loadmat mat = loadmat(mat_path) 然后根据你的需要将mat里的信息保存就行了。

[jone222]

是直接在train中加入这两行代码吗

[mpmmpmmmp]

或许对你有帮助: https://github.com/JiajiaLi04/Soybean-Pod-Counting-from-UAV-Images

[jone222]

我看一下，感谢

[willow-god]

I want to change backbone to ResNet, how should I modify the code我想将骨干更改为ResNet，我应该如何修改代码

Modify backbone.py to return four feature layers. Of course, only the last three feature layers are used.

修改backbone.py返回四个特征层，当然只使用最后三个特征层。

你好大佬，我也尝试修改backbone，现在代码运行了，但是结果不是很理想，我使用了resnet的四层特征，并在decoder中添加了一层进行处理，不过效果反而下降了，我不清楚是什么原因，请问佬有什么想法吗？

[KRK11]

I want to change backbone to ResNet, how should I modify the code我想将骨干更改为ResNet，我应该如何修改代码

Modify backbone.py to return four feature layers. Of course, only the last three feature layers are used. 修改backbone.py返回四个特征层，当然只使用最后三个特征层。

你好大佬，我也尝试修改backbone，现在代码运行了，但是结果不是很理想，我使用了resnet的四层特征，并在decoder中添加了一层进行处理，不过效果反而下降了，我不清楚是什么原因，请问佬有什么想法吗？

效果差大不大？

[willow-god]

I want to change backbone to ResNet, how should I modify the code我想将骨干更改为ResNet，我应该如何修改代码

Modify backbone.py to return four feature layers. Of course, only the last three feature layers are used. 修改backbone.py返回四个特征层，当然只使用最后三个特征层。

你好大佬，我也尝试修改backbone，现在代码运行了，但是结果不是很理想，我使用了resnet的四层特征，并在decoder中添加了一层进行处理，不过效果反而下降了，我不清楚是什么原因，请问佬有什么想法吗？

效果差大不大？

差的不是很多，但是我感觉不应该下降VGG换ResNet感觉怎么着也得提升吧😂😂😂，模型大致修改了这些位置：

上面这个是基线网络的特征提取

上面这个是decoder加了一层