listing the detection and recognition models not working in pytorch

[nikokks]

Bug description

Hi,

This is the list of models not working in inference:

• detection models:

  • db_resnet_34
  • linknet_resnet18
  • linknet_resnet34
  • linknet_resnet50 for all these detection models in pytorch, the code inference works but it get to much segmentations/detected boxes.

• recognition models:

  • sar_resnet31
  • master
  • vitstr_small
  • vitstr_base For all of these models in pytorch, the code inference seems to work, but it gets random character recognition. I think they are not correct loaded or trained.

Code snippet to reproduce the bug

in doctr/demo/backend/pytorch.py file, changing this lines

DET_ARCHS = ["db_resnet50", "db_mobilenet_v3_large", "linknet_resnet50_rotation"]
RECO_ARCHS = ["crnn_vgg16_bn", "crnn_mobilenet_v3_small", "master", "sar_resnet31"]

by this lines

#DET_ARCHS = ["db_resnet50", "db_mobilenet_v3_large", "linknet_resnet50_rotation"]
DET_ARCHS = [ "db_resnet34",
        "db_resnet50",
        "db_mobilenet_v3_large",
        "linknet_resnet18",
        "linknet_resnet34",
        "linknet_resnet50",
        "db_resnet50_rotation"]
#RECO_ARCHS = ["crnn_vgg16_bn", "crnn_mobilenet_v3_small", "master", "sar_resnet31"]
RECO_ARCHS=[ "crnn_vgg16_bn",
    "crnn_mobilenet_v3_small",
    "crnn_mobilenet_v3_large",
    "sar_resnet31",
    "master",
    "vitstr_small",
    "vitstr_base"]

and running this code to try all the pytorch models

USE_TF=0 streamlit run demo/app.py

Error traceback

not errors but bugs on models

Environment

wget https://raw.githubusercontent.com/mindee/doctr/main/scripts/collect_env.py

For security purposes, please check the contents of collect_env.py before running it.

python collect_env.py

Deep Learning backend

from doctr.file_utils import is_tf_available, is_torch_available

print(f"is_tf_available: {is_tf_available()}") print(f"is_torch_available: {is_torch_available()}")

[felixdittrich92]

Hey @nikokks :wave:,

the most of this models does not have pretrained checkpoints but @odulcy-mindee and @charlesmindee are currently on it so stay tuned #969

[odulcy-mindee]

Hello @nikokks, Thanks for this issue! Indeed, we're on it, we'll release model one by one when they'll be trained :+1: