Python virtual machine crashes with segfault

[CdAB63]

Good night.

EasyOCR is crashing the Python 3.10 VM Ubuntu 22.04 / Debian Bullseye (identical problem but Python 3.9) Architecture Raspberry Pi 4 4GBytes

Error as follows:

Python 3.10.12 (main, Jun 11 2023, 05:26:28) [GCC 11.4.0] on linux Type "help", "copyright", "credits" or "license" for more information.

import faulthandler import easyocr faulthandler.enable() reader = easyocr.Reader(['pt']) Neither CUDA nor MPS are available - defaulting to CPU. Note: This module is much faster with a GPU. Fatal Python error: Segmentation fault

Current thread 0x0000ffff949d3420 (most recent call first): File "/home/cdab63/.local/lib/python3.10/site-packages/torch/_ops.py", line 692 in call File "/home/cdab63/.local/lib/python3.10/site-packages/torch/ao/nn/quantized/dynamic/modules/rnn.py", line 119 in init File "/home/cdab63/.local/lib/python3.10/site-packages/torch/ao/nn/quantized/dynamic/modules/rnn.py", line 400 in init File "/home/cdab63/.local/lib/python3.10/site-packages/torch/ao/nn/quantized/dynamic/modules/rnn.py", line 291 in from_float File "/home/cdab63/.local/lib/python3.10/site-packages/torch/ao/nn/quantized/dynamic/modules/rnn.py", line 495 in from_float File "/home/cdab63/.local/lib/python3.10/site-packages/torch/ao/quantization/quantize.py", line 623 in swap_module File "/home/cdab63/.local/lib/python3.10/site-packages/torch/ao/quantization/quantize.py", line 590 in _convert File "/home/cdab63/.local/lib/python3.10/site-packages/torch/ao/quantization/quantize.py", line 588 in _convert File "/home/cdab63/.local/lib/python3.10/site-packages/torch/ao/quantization/quantize.py", line 588 in _convert File "/home/cdab63/.local/lib/python3.10/site-packages/torch/ao/quantization/quantize.py", line 550 in convert File "/home/cdab63/.local/lib/python3.10/site-packages/torch/ao/quantization/quantize.py", line 465 in quantize_dynamic File "/home/cdab63/.local/lib/python3.10/site-packages/easyocr/recognition.py", line 177 in get_recognizer File "/home/cdab63/.local/lib/python3.10/site-packages/easyocr/easyocr.py", line 231 in init File "", line 1 in

Extension modules: PIL._imaging, numpy.core._multiarray_umath, numpy.core._multiarray_tests, numpy.linalg._umath_linalg, numpy.fft._pocketfft_internal, numpy.random._common, numpy.random.bit_generator, numpy.random._bounded_integers, numpy.random._mt19937, numpy.random.mtrand, numpy.random._philox, numpy.random._pcg64, numpy.random._sfc64, numpy.random._generator, torch._C, torch._C._fft, torch._C._linalg, torch._C._nested, torch._C._nn, torch._C._sparse, torch._C._special, PIL._imagingft, cv2.cv2, scipy._lib._ccallback_c, scipy.ndimage._nd_image, scipy.special._ufuncs_cxx, scipy.special._ufuncs, scipy.special._specfun, scipy.special._comb, scipy.linalg._fblas, scipy.linalg._flapack, scipy.linalg.cython_lapack, scipy.linalg._cythonized_array_utils, scipy.linalg._solve_toeplitz, scipy.linalg._decomp_lu_cython, scipy.linalg._matfuncs_sqrtm_triu, scipy.linalg.cython_blas, scipy.linalg._matfuncs_expm, scipy.linalg._decomp_update, scipy.sparse._sparsetools, _csparsetools, scipy.sparse._csparsetools, scipy.sparse.linalg._isolve._iterative, scipy.sparse.linalg._dsolve._superlu, scipy.sparse.linalg._eigen.arpack._arpack, scipy.sparse.csgraph._tools, scipy.sparse.csgraph._shortest_path, scipy.sparse.csgraph._traversal, scipy.sparse.csgraph._min_spanning_tree, scipy.sparse.csgraph._flow, scipy.sparse.csgraph._matching, scipy.sparse.csgraph._reordering, scipy.linalg._flinalg, scipy.special._ellip_harm_2, _ni_label, scipy.ndimage._ni_label, skimage._shared.geometry, yaml._yaml (total: 58) Segmentation fault (core dumped)

[CdAB63]

Id does not seem to be related to opencv but to torch (where the fail happens)

[CdAB63]

Fixed the problem retrofitting torch and torchvision

Python 3.10.12 (main, Jun 11 2023, 05:26:28) [GCC 11.4.0] on linux Type "help", "copyright", "credits" or "license" for more information.

import torch print(torch.version) 2.0.1 import torchvision print(torchvision.version) 0.15.2 import easyocr reader = easyocr.Reader(['pt']) Neither CUDA nor MPS are available - defaulting to CPU. Note: This module is much faster with a GPU.

[manthey]

I've noticed that on a cpu and current versions of torch you get a segfault, but this can be avoided by passing quantize=False when instantiating the Reader. Specifically, it looks like this line: https://github.com/JaidedAI/EasyOCR/blob/master/easyocr/recognition.py#L177 doesn't pass through the device (e.g., I think (but am not certain), that it should be torch.quantization.quantize_dynamic(model, {'device': device}, dtype=torch.qint8, inplace=True)).

But I don't see quantize in the official docs, so I'm not sure what implication this has.

And, in looking at this, I wonder why this line https://github.com/JaidedAI/EasyOCR/blob/master/easyocr/easyocr.py#L89 has a comma at the end. It seems like it makes self.quantize a tuple.

[Shoz0]

I've noticed that on a cpu and current versions of torch you get a segfault, but this can be avoided by passing quantize=False when instantiating the Reader. Specifically, it looks like this line: https://github.com/JaidedAI/EasyOCR/blob/master/easyocr/recognition.py#L177 doesn't pass through the device (e.g., I think (but am not certain), that it should be torch.quantization.quantize_dynamic(model, {'device': device}, dtype=torch.qint8, inplace=True)).

But I don't see quantize in the official docs, so I'm not sure what implication this has.

And, in looking at this, I wonder why this line https://github.com/JaidedAI/EasyOCR/blob/master/easyocr/easyocr.py#L89 has a comma at the end. It seems like it makes self.quantize a tuple.

Thank you so much! Adding 'quantize=False' when instantiating the Reader, solve the problem of crash when using reader.readtext() for me.

reader = easyocr.Reader(['en'], gpu=False, quantize=False)