You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
try quantize=False, if that stops the crash then we can use a newer version of pytorch. If we use newer pytorch then we can use newer easyocr. If we use newer easyocr then we can use newer Pillow. Still can't use numpy2 though.
ensure quantize=False doesn't degrade the quality of detection
quantize=False might make it slower (test on both CPU and GPU)
The text was updated successfully, but these errors were encountered:
Problems with easyocr:
easyocr is tied to 1.6.2 because later versions need
torch.backends.mps
which only comes in torch 1.12easyocr < 1.7 needs Pillow <10.0
easyocr is tied to numpy<2, see numpy 2 JaidedAI/EasyOCR#1301, no word on numpy2 support yet
Crashes with illegal instruction, need to use older version of pytorch, see
Illegal instruction (core dumped) JaidedAI/EasyOCR#1174 (comment)
max pytorch version is torch==1.11.0, torchvision==0.12.0
quantize=False
parameter toeasyocr.Reader()
, see Python virtual machine crashes with segfault JaidedAI/EasyOCR#1155To do:
The text was updated successfully, but these errors were encountered: