When I use my own dataset to train the BEIT network, there is no problem with the training process, but the following error is reported during validation, what is the reason?

[vGuiGui6]

11/12 04:39:27 - mmengine - INFO - Exp name: beit-base_upernet_8xb2-160k_parcel20k-640x640_20241111_235427
11/12 04:39:27 - mmengine - INFO - Iter(train) [ 2000/20000] base_lr: 2.9906e-05 lr: 7.6016e-06 eta: 1 day, 18:42:25 time: 8.5668 data_time: 0.0020 memory: 11904 loss: 0.4703 decode.loss_ce: 0.3251 decode.acc_seg: 75.8877 aux.loss_ce: 0.1451 aux.acc_seg: 76.0485
11/12 04:39:27 - mmengine - INFO - Saving checkpoint at 2000 iterations
Traceback (most recent call last):
File "E:\Net\mmsegmentation\tools\train.py", line 106, in
main()
File "E:\Net\mmsegmentation\tools\train.py", line 102, in main
runner.train()
File "D:\ProgramData\Anaconda3\envs\open-mmlab\lib\site-packages\mmengine\runner\runner.py", line 1777, in train
model = self.train_loop.run() # type: ignore
File "D:\ProgramData\Anaconda3\envs\open-mmlab\lib\site-packages\mmengine\runner\loops.py", line 294, in run
self.runner.val_loop.run()
File "D:\ProgramData\Anaconda3\envs\open-mmlab\lib\site-packages\mmengine\runner\loops.py", line 373, in run
self.run_iter(idx, data_batch)
File "D:\ProgramData\Anaconda3\envs\open-mmlab\lib\site-packages\torch\autograd\grad_mode.py", line 27, in decorate_context
return func(*args, **kwargs)
File "D:\ProgramData\Anaconda3\envs\open-mmlab\lib\site-packages\mmengine\runner\loops.py", line 393, in run_iter
outputs = self.runner.model.val_step(data_batch)
File "D:\ProgramData\Anaconda3\envs\open-mmlab\lib\site-packages\mmengine\model\base_model\base_model.py", line 133, in val_step
return self._run_forward(data, mode='predict') # type: ignore
File "D:\ProgramData\Anaconda3\envs\open-mmlab\lib\site-packages\mmengine\model\base_model\base_model.py", line 361, in _run_forward
results = self(**data, mode=mode)
File "D:\ProgramData\Anaconda3\envs\open-mmlab\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl
return forward_call(*input, **kwargs)
File "E:\Net\mmsegmentation\mmseg\models\segmentors\base.py", line 96, in forward
return self.predict(inputs, data_samples)
File "E:\Net\mmsegmentation\mmseg\models\segmentors\encoder_decoder.py", line 220, in predict
seg_logits = self.inference(inputs, batch_img_metas)
File "E:\Net\mmsegmentation\mmseg\models\segmentors\encoder_decoder.py", line 341, in inference
seg_logit = self.slide_inference(inputs, batch_img_metas)
File "E:\Net\mmsegmentation\mmseg\models\segmentors\encoder_decoder.py", line 283, in slide_inference
crop_seg_logit = self.encode_decode(crop_img, batch_img_metas)
File "E:\Net\mmsegmentation\mmseg\models\segmentors\encoder_decoder.py", line 128, in encode_decode
x = self.extract_feat(inputs)
File "E:\Net\mmsegmentation\mmseg\models\segmentors\encoder_decoder.py", line 119, in extract_feat
x = self.backbone(inputs)
File "D:\ProgramData\Anaconda3\envs\open-mmlab\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl
return forward_call(*input, **kwargs)
File "E:\Net\mmsegmentation\mmseg\models\backbones\beit.py", line 535, in forward
x = layer(x)
File "D:\ProgramData\Anaconda3\envs\open-mmlab\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl
return forward_call(*input, **kwargs)
File "E:\Net\mmsegmentation\mmseg\models\backbones\beit.py", line 221, in forward
x = x + self.drop_path(self.gamma_1 * self.attn(self.norm1(x)))
File "D:\ProgramData\Anaconda3\envs\open-mmlab\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl
return forward_call(*input, **kwargs)
File "E:\Net\mmsegmentation\mmseg\models\backbones\beit.py", line 142, in forward
attn = attn + relative_position_bias.unsqueeze(0)
RuntimeError: The size of tensor a (1025) must match the size of tensor b (1601) at non-singleton dimension 3

Process finished with exit code 1

[roiyfox]

Due to the separation of the training and test sets, their cropping resolutions are different. As a result, your convolution cannot evenly divide the images in the test set, leading to tensor concatenation errors. You can try adding size_divisor=32 to the data_preprocessor section in your configuration file for the test set, which might solve your problem. I encountered a similar issue before.