main/version3.x/module_usage/table_structure_recognition

[giscus[bot]]

main/version3.x/module_usage/table_structure_recognition

Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and deployment among server, mobile, embedded and IoT devices)

https://www.paddleocr.ai/main/version3.x/module_usage/table_structure_recognition.html

[lyj201644070230]

表格数据集是用什么标注的？ PPlabelOCR吗？

[lyj201644070230]

SLANeXt_wired 为什么训练不起来？形状不对导致报错。Traceback (most recent call last):
File "/data8T/yajieliu/PaddleOCR-3.1.0/tools/train.py", line 272, in
main(config, device, logger, vdl_writer, seed)
File "/data8T/yajieliu/PaddleOCR-3.1.0/tools/train.py", line 225, in main
program.train(
File "/data8T/yajieliu/PaddleOCR-3.1.0/tools/program.py", line 356, in train
preds = model(images, data=batch[1:])
File "/home/hnxt/anaconda3/envs/yajieliu-ppocr-3.1/lib/python3.9/site-packages/paddle/nn/layer/layers.py", line 1571, in call
return self.forward(*inputs, **kwargs)
File "/home/hnxt/anaconda3/envs/yajieliu-ppocr-3.1/lib/python3.9/site-packages/paddle/distributed/parallel.py", line 570, in forward
outputs = self._layers(*inputs, **kwargs)
File "/home/hnxt/anaconda3/envs/yajieliu-ppocr-3.1/lib/python3.9/site-packages/paddle/nn/layer/layers.py", line 1571, in call
return self.forward(*inputs, **kwargs)
File "/data8T/yajieliu/PaddleOCR-3.1.0/ppocr/modeling/architectures/base_model.py", line 85, in forward
x = self.backbone(x)
File "/home/hnxt/anaconda3/envs/yajieliu-ppocr-3.1/lib/python3.9/site-packages/paddle/nn/layer/layers.py", line 1571, in call
return self.forward(*inputs, **kwargs)
File "/data8T/yajieliu/PaddleOCR-3.1.0/ppocr/modeling/backbones/rec_vary_vit.py", line 550, in forward
cnn_feature = self.vision_tower_high(pixel_values)
File "/home/hnxt/anaconda3/envs/yajieliu-ppocr-3.1/lib/python3.9/site-packages/paddle/nn/layer/layers.py", line 1571, in call
return self.forward(*inputs, **kwargs)
File "/data8T/yajieliu/PaddleOCR-3.1.0/ppocr/modeling/backbones/rec_vary_vit.py", line 180, in forward
x = blk(x)
File "/home/hnxt/anaconda3/envs/yajieliu-ppocr-3.1/lib/python3.9/site-packages/paddle/nn/layer/layers.py", line 1571, in call
return self.forward(*inputs, **kwargs)
File "/data8T/yajieliu/PaddleOCR-3.1.0/ppocr/modeling/backbones/rec_vary_vit.py", line 245, in forward
x = self.attn(x)
File "/home/hnxt/anaconda3/envs/yajieliu-ppocr-3.1/lib/python3.9/site-packages/paddle/nn/layer/layers.py", line 1571, in call
return self.forward(*inputs, **kwargs)
File "/data8T/yajieliu/PaddleOCR-3.1.0/ppocr/modeling/backbones/rec_vary_vit.py", line 312, in forward
attn = add_decomposed_rel_pos(
File "/data8T/yajieliu/PaddleOCR-3.1.0/ppocr/modeling/backbones/rec_vary_vit.py", line 446, in add_decomposed_rel_pos
attn.reshape([B, q_h, q_w, k_h, k_w])
TypeError: unsupported operand type(s) for +: 'Tensor' and 'Tensor'
I0813 16:53:51.677794 2296949 process_group_nccl.cc:159] ProcessGroupNCCL destruct
I0813 16:53:51.707840 2297032 tcp_store.cc:292] receive shutdown event and so quit from MasterDaemon run loop
LAUNCH INFO 2025-08-13 16:53:52,484 Exit code 1 并且官方提取的预训练权重SLANeXt_wired_pretrained.pdparams的pos_embed 形状为 [1, 32, 32, 768]

[HK-SOURCE-beep]

兄弟解决了吗，遇到同样的问题。使用ppstructure的predict_table运行不了SLANeXt和SLANeXt_wired，SLANet和SLANetPlus倒没这个问题

[HK-SOURCE-beep]

你好，用OCRv5运行ppstruture的predict_table.py报错，用OCRv4就没事，请问这是咋回事，谢谢！
Traceback (most recent call last):
File "/root/autodl-tmp/PaddleOCR/ppstructure/table/predict_table.py", line 241, in
main(args)
File "/root/autodl-tmp/PaddleOCR/ppstructure/table/predict_table.py", line 189, in main
pred_res, _ = table_sys(img)
File "/root/autodl-tmp/PaddleOCR/ppstructure/table/predict_table.py", line 100, in call
dt_boxes, rec_res, det_elapse, rec_elapse = self._ocr(copy.deepcopy(img))
File "/root/autodl-tmp/PaddleOCR/ppstructure/table/predict_table.py", line 145, in _ocr
rec_res, rec_elapse = self.text_recognizer(img_crop_list)
File "/root/autodl-tmp/PaddleOCR/tools/infer/predict_rec.py", line 826, in call
rec_result = self.postprocess_op(
File "/root/autodl-tmp/PaddleOCR/ppocr/postprocess/rec_postprocess.py", line 208, in call
text = self.decode(
File "/root/autodl-tmp/PaddleOCR/ppocr/postprocess/rec_postprocess.py", line 156, in decode
char_list = [
File "/root/autodl-tmp/PaddleOCR/ppocr/postprocess/rec_postprocess.py", line 157, in
self.character[text_id] for text_id in text_index[batch_idx][selection]
IndexError: list index out of range

[HK-SOURCE-beep]

和楼上问题类似，使用SLANeXt_wired显示尺寸对不上，下面是报错
Traceback (most recent call last):
File "/root/autodl-tmp/PaddleOCR/ppstructure/table/predict_table.py", line 241, in
main(args)
File "/root/autodl-tmp/PaddleOCR/ppstructure/table/predict_table.py", line 189, in main
pred_res, _ = table_sys(img)
File "/root/autodl-tmp/PaddleOCR/ppstructure/table/predict_table.py", line 96, in call
structure_res, elapse = self._structure(copy.deepcopy(img))
File "/root/autodl-tmp/PaddleOCR/ppstructure/table/predict_table.py", line 118, in _structure
structure_res, elapse = self.table_structurer(copy.deepcopy(img))
File "/root/autodl-tmp/PaddleOCR/ppstructure/table/predict_structure.py", line 140, in call
self.predictor.run()
ValueError: In user code:

InvalidArgumentError: Broadcast dimension mismatch. Operands could not be broadcast together with the shape of X = [1, 30, 30, 768] and the shape of Y = [1, 32, 32, 768]. Received [30] in X is not equal to [32] in Y at i:1.
  [Hint: Expected x_dims_array[i] == y_dims_array[i] || (x_dims_array[i] <= 1 && x_dims_array[i] != 0) || (y_dims_array[i] <= 1 && y_dims_array[i] != 0) == true, but received x_dims_array[i] == y_dims_array[i] || (x_dims_array[i] <= 1 && x_dims_array[i] != 0) || (y_dims_array[i] <= 1 && y_dims_array[i] != 0):0 != true:1.] (at ../paddle/phi/kernels/funcs/common_shape.h:72)
  [operator < pd_kernel.phi_kernel > error]

[HK-SOURCE-beep]

这两个表格结构识别器可以单独使用，但使用ppstructure/predict_table和OCR配合使用就报错。SLANet和SLANetPlus就不报错

[lyj201644070230]

我目前也是SLANeXt_wired尺寸对不上，暂时不知道怎么解决