NRTR Export mismatch dim

[leduy-it]

🔎 Search before asking

• I have searched the PaddleOCR Docs and found no similar bug report.
• I have searched the PaddleOCR Issues and found no similar bug report.
• I have searched the PaddleOCR Discussions and found no similar bug report.

🐛 Bug (问题描述)

I'm trying export and using NRTR model follow doc, but i encounter, i'm using paddlepaddle-gpu==2.6.2
Reproduce just download the weight and do the same follow the doc:
https://paddlepaddle.github.io/PaddleOCR/main/en/algorithm/text_recognition/algorithm_rec_nrtr.html#41-python-inference

python3 tools/infer/predict_rec.py --image_dir='./image/12.jpg' --rec_model_dir='./inference/rec_mtb_nrtr/' --rec_algorithm='NRTR' --rec_image_shape='1,32,100' --rec_char_dict_path='./ppocr/utils/EN_symbol_dict.txt'
grep: warning: GREP_OPTIONS is deprecated; please use an alias or script
[2025/04/28 11:43:18] ppocr WARNING: The first GPU is used for inference by default, GPU ID: 0
[2025/04/28 11:43:20] ppocr INFO: In PP-OCRv3, rec_image_shape parameter defaults to '3, 48, 320', if you are using recognition model with PP-OCRv2 or an older version, please set --rec_image_shape='3,32,320
norm_img_batch shape before inference: (1, 1, 32, 100)
[2025/04/28 11:43:21] ppocr INFO: Traceback (most recent call last):
File "/workspace/JP_FreeForm/tools/infer/predict_rec.py", line 937, in main
rec_res, _ = text_recognizer(img_list)
File "/workspace/JP_FreeForm/tools/infer/predict_rec.py", line 870, in call
self.predictor.run()
ValueError: (InvalidArgument) The 'shape' in ReshapeOp is invalid. The input tensor X'size must be equal to the capacity of 'shape'. But received X's shape = [1, 2, 1536], X's size = 3072, 'shape' is [1, 1, 3, 8, 64], the capacity of 'shape' is 1536.
[Hint: Expected capacity == in_size, but received capacity:1536 != in_size:3072.] (at /paddle/paddle/phi/infermeta/unary.cc:1955)
[operator < reshape2 > error]

[2025/04/28 11:43:21] ppocr INFO: (InvalidArgument) The 'shape' in ReshapeOp is invalid. The input tensor X'size must be equal to the capacity of 'shape'. But received X's shape = [1, 2, 1536], X's size = 3072, 'shape' is [1, 1, 3, 8, 64], the capacity of 'shape' is 1536.
[Hint: Expected capacity == in_size, but received capacity:1536 != in_size:3072.] (at /paddle/paddle/phi/infermeta/unary.cc:1955)
[operator < reshape2 > error]

Thanks in advanced!

🏃‍♂️ Environment (运行环境)

ubtuntu
paddlepaddle-gpu==2.6.2
latest paddleocr

🌰 Minimal Reproducible Example (最小可复现问题的Demo)

python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png' --rec_model_dir='./inference/rec_mtb_nrtr/' --rec_algorithm='NRTR' --rec_image_shape='1,32,100' --rec_char_dict_path='./ppocr/utils/EN_symbol_dict.txt'

[cuicheng01]

Hello, could you please try using paddle version 3.0? If you encounter any detailed error messages, please let us know. Thank you!

[leduy-it]

@cuicheng01 Thanks for your quickly reply. Yes, i have try version 3.0 and still the same error was raised.
Can you take a look for this algothrim because it's has been released quite a long time and need some new documents.
Hope some good new from u!
Thanks.

[leduy-it]

@cuicheng01 More information that i can training and evaluation with pdparams smoothly but when i wanna export it to static pdinference, i counter this error.
@Topdu and Topdu i know you are the author of this algorithm, can you give me some help or at least advice, thanks in advance!