[General Issue]train contrib/PanopticDeepLab时出现Logit和Label的shape不同报错

[MorphlingX]

2022-08-10 02:33:20 [INFO]
------------Environment Information------------- platform: Windows-10-10.0.18362-SP0 Python: 3.8.9 (tags/v3.8.9:a743f81, Apr 6 2021, 14:02:34) [MSC v.1928 64 bit (AMD64)] Paddle compiled with cuda: True NVCC: Cuda compilation tools, release 10.2, V10.2.89 cudnn: 7.6 GPUs used: 1 CUDA_VISIBLE_DEVICES: None GPU: ['GPU 0: NVIDIA GeForce'] PaddleSeg: 2.6.0 PaddlePaddle: 2.3.1 OpenCV: 4.6.0

2022-08-10 02:33:22 [INFO]
---------------Config Information--------------- batch_size: 2 iters: 90000 learning_rate: decay: end_lr: 0.0 power: 0.9 type: poly value: 1.25e-05 loss: coef:

• 1
• 200
• 0.001 types:
• ignore_index: 255 top_k_percent_pixels: 0.2 type: CrossEntropyLoss
• ignore_index: 255 reduction: none type: MSELoss
• ignore_index: 255 reduction: none type: L1Loss model: align_corners: true aspp_out_channels: 256 aspp_ratios:
• 1
• 3
• 6
• 9 backbone: output_stride: 32 pretrained: https://bj.bcebos.com/paddleseg/dygraph/resnet50_vd_ssld_v2.tar.gz type: ResNet50_vd backbone_indices:
• 2
• 1
• 0
• 3 decoder_channels: 256 instance_aspp_out_channels: 256 instance_class_key:
• center
• offset instance_decoder_channels: 128 instance_head_channels: 32 instance_low_level_channels_projects:
• 64
• 32
• 16 instance_num_classes:
• 1
• 2 low_level_channels_projects:
• 128
• 64
• 32 type: PanopticDeepLab optimizer: type: adam train_dataset: dataset_root: data/cityscapes ignore_stuff_in_offset: true mode: train small_instance_area: 4096 small_instance_weight: 3 transforms:
• max_scale_factor: 2.0 min_scale_factor: 0.5 scale_step_size: 0.25 type: ResizeStepScaling
• crop_size:
  • 1025
  • 513 label_padding_value:
  • 0
  • 0
  • 0 type: RandomPaddingCrop
• type: RandomHorizontalFlip
• brightness_range: 0.4 contrast_range: 0.4 saturation_range: 0.4 type: RandomDistort
• type: Normalize type: CityscapesPanoptic val_dataset: dataset_root: data/cityscapes ignore_stuff_in_offset: true mode: val small_instance_area: 4096 small_instance_weight: 3 transforms:
• label_padding_value:
  • 0
  • 0
  • 0 target_size:
  • 2049
  • 1025 type: Padding
• type: Normalize type: CityscapesPanoptic

---

W0810 02:33:24.015547 6100 gpu_resources.cc:61] Please NOTE: device: 0, GPU Compute Capability: 7.5, Driver API Version: 11.7, Runtime API Version: 10.2 W0810 02:33:24.015547 6100 gpu_resources.cc:91] device: 0, cuDNN Version: 7.6. 2022-08-10 02:33:24 [INFO] Loading pretrained model from https://bj.bcebos.com/paddleseg/dygraph/resnet50_vd_ssld_v2.tar.gz 2022-08-10 02:33:25 [INFO] There are 275/275 variables loaded into ResNet_vd. 2022-08-10 02:33:25 [WARNING] learning_rate in configuration file will be deprecated, please use lr_scheduler instead. E.g lr_scheduler: type: PolynomialDecay learning_rate: 0.01 C:\Users\LE\AppData\Local\Programs\Python\Python38\lib\site-packages\paddle\nn\layer\norm.py:653: UserWarning: When training, we now always track global mean and variance. warnings.warn( Traceback (most recent call last): File "train.py", line 177, in main(args) File "train.py", line 155, in main train( File "D:\PaddleSeg\contrib\PanopticDeepLab\core\train.py", line 174, in train loss_list = loss_computation( File "D:\PaddleSeg\contrib\PanopticDeepLab\core\train.py", line 39, in loss_computation semantic_loss = losses['types'][0](logits_list[0], semantic, File "C:\Users\LE\AppData\Local\Programs\Python\Python38\lib\site-packages\paddle\fluid\dygraph\layers.py", line 930, in call return self._dygraph_call_func(*inputs, **kwargs) File "C:\Users\LE\AppData\Local\Programs\Python\Python38\lib\site-packages\paddle\fluid\dygraph\layers.py", line 915, in _dygraph_call_func outputs = self.forward(*inputs, **kwargs) File "d:\paddleseg\paddleseg\models\losses\cross_entropy_loss.py", line 82, in forward loss = F.cross_entropy( File "C:\Users\LE\AppData\Local\Programs\Python\Python38\lib\site-packages\paddle\nn\functional\loss.py", line 1735, in cross_entropy _, out = _C_ops.softmax_with_cross_entropy( ValueError: (InvalidArgument) Input(Logits) and Input(Label) should in same shape in dimensions except axis. [Hint: Expected logits_dims[i] == labels_dims[i], but received logits_dims[i]:513 != labels_dims[i]:1024.] (at ..\paddle/fluid/operators/softmax_with_cross_entropy_op.cc:164)
[operator < softmax_with_cross_entropy > error]

Thanks for your issue. To help us solve the issue better, please provide following information：

1. PaddleSeg version:2.6.0 (please specify the branch as well，e.g. PaddleSeg release/2.3)
2. PaddlePaddle version: 2.3.1(e.g. PaddlePaddle 2.1.0)
3. Operation system:Windows10 (e.g. Linux/Windows/MacOS)
4. Python version: 3.8(e.g. Python3.6/7/8)
5. CUDA/cuDNN version: CUDA10.2/cuDNN 7.6.5(e.g. CUDA10.2/cuDNN 7.6.5)
6. Additional context: (Add any other context about the problem)

---

欢迎您反馈PaddleSeg使用问题，辛苦您提供以下信息，方便我们快速定位和解决问题：

1. PaddleSeg版本：2.6.0（请提供版本号和分支信息，如PaddleSeg release/2.3）
2. PaddlePaddle版本：2.3.1（如PaddlePaddle 2.1.0）
3. 操作系统信息：Windows10（如Linux/Windows/MacOS）
4. Python版本号：3.8（如Python3.6/7/8）
5. CUDA/cuDNN版本：CUDA10.2/cuDNN7.6.5（ 如CUDA10.2/cuDNN 7.6.5等）
6. 其他内容: （增加其他与问题相关的内容）

[MorphlingX]

将PaddleSeg/contribPanopticDeepLab/datasets/cityscapes_panoptic.py中161行im, label = self.transforms(im=image_path, label=label_path)（运行时报错im参数不存在）改成了

tmp = self.transforms(dict(img=image_path, label=label_path))
im = tmp['img']
label = tmp['label']

可能导致了问题？

[wuyefeilin]

新版本paddleseg的transform进行了升级，需要对应升级一下，后面会进行更新。

你可以先使用2.5版本。看看是否能正常运行

[MorphlingX]

新版本paddleseg的transform进行了升级，需要对应升级一下，后面会进行更新。

你可以先使用2.5版本。看看是否能正常运行

已经能够正常运行，单卡学习学习率需要改为1/4吗

[wuyefeilin]

新版本已经修复该问题了。

学习率问题不一定需要修改，你可以做一下对比实验