OCR predictor returns empty detection in docTR 0.11.0

[AmineM24]

Bug description

I've encountered an issue after upgrading to docTR version 0.11.0. When I run the OCR predictor on images that previously worked fine with version 0.8.1, I now get empty detection results.

Could this be a regression or something related to default configurations introduced in recent updates? Any insights or guidance would be appreciated

Thanks a lot

Code snippet to reproduce the bug

from PIL import Image
import numpy as np

model = ocr_predictor(pretrained=True)

img = Image.open("my_image.png").convert("RGB")
arr = np.array(img)
result = model([arr]).export()
print(result)

The exact same code and image provide normal OCR results (detected blocks of text) when downgrading back to version 0.8.1.

Error traceback

I don't get any error messages, but instead, the result contains no detected text blocks

{'pages': [{'page_idx': 0, 'dimensions': (1600, 1131), 'orientation': {'value': None, 'confidence': None}, 'language': {'value': None, 'confidence': None}, 'blocks': []}]

Environment

DocTR version: v0.11.0
TensorFlow version: 2.18.0
PyTorch version: 2.5.1+cu124 (torchvision 0.20.1+cu124)
OpenCV version: 4.11.0
OS: Ubuntu 22.04.4 LTS
Python version: 3.11.11
Is CUDA available (TensorFlow): Yes
Is CUDA available (PyTorch): Yes
CUDA runtime version: 12.5.82
GPU models and configuration: GPU 0: Tesla T4
Nvidia driver version: 550.54.15
cuDNN version: Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.9.2.1
/usr/lib/x86_64-linux-gnu/libcudnn_adv.so.9.2.1
/usr/lib/x86_64-linux-gnu/libcudnn_cnn.so.9.2.1
/usr/lib/x86_64-linux-gnu/libcudnn_engines_precompiled.so.9.2.1
/usr/lib/x86_64-linux-gnu/libcudnn_engines_runtime_compiled.so.9.2.1
/usr/lib/x86_64-linux-gnu/libcudnn_graph.so.9.2.1
/usr/lib/x86_64-linux-gnu/libcudnn_heuristic.so.9.2.1
/usr/lib/x86_64-linux-gnu/libcudnn_ops.so.9.2.1

Deep Learning backend

is_tf_available: True
is_torch_available: True

[felixdittrich92]

Hi @AmineM24 👋,

Seems to be related to #1884

If you have both backends (TF and PyTorch) in the same env installed please use the USE_TF= and USE_TORCH= env vars ..there is any conflicting dependency between PyTorch and TensorFlow

[felixdittrich92]

Any update @AmineM24 ? :)