Skip to content

Latest commit

 

History

History
37 lines (24 loc) · 2.54 KB

TestPretrainedModel.md

File metadata and controls

37 lines (24 loc) · 2.54 KB
Raw

Table Detection Task

This provides a pipeline to test pretrained model and visualize the performance of Table Detection task.

Table detection aims to locate tables using bounding boxes in a document. Given a document page in the image format, generating several bounding box that represents the location of tables in this page.

The authors pretrained two models (ResNeXt-101 and ResNeXt-152) using the Detectron library. So we need to install Detectron first.

The installation of Detectron is introduced officially here. Detectron is based on Caffe2, which is now integrated into Pytorch library now. But be careful that the operators in detectron has no CPU version, so a GPU system is needed.. Also, when I'm installing the dependencies according to https://github.com/facebookresearch/Detectron/blob/master/INSTALL.md#detectron by pip install -r $DETECTRON/requirements.txt, I notice that I must use pip install pyyaml==3.12 to install a elder version of pyyaml instead of the latest, or I will meet the error BBOX_XFORM_CLIP: !!python/object/apply:numpy.core. This issue is posted here.

The complete commands are

# DETECTRON=/path/to/clone/detectron
git clone https://github.com/facebookresearch/detectron $DETECTRON
pip install pyyaml==3.12
pip install -r $DETECTRON/requirements.txt
cd $DETECTRON && make
python $DETECTRON/detectron/tests/test_spatial_narrow_as_op.py

I will experiment on the ResNeXt101 model in the following, while the pipeline for ResNeXt152 is similiar.

You need to download the configuration file and the weights of the model first here. In the following, we use $MODEL_PATH to denote the downloaded weights and $CONFIG_MODEL to denote the configuration file.

Then simply run the commands

python tools/infer_simple.py --cfg $CONFIG_MODEL --output-dir /tmp/detectron-tablebank --image-ext jpg \
    --wts $MODEL_PATH /home/shr/TableBank/data/Sampled_Detection_data/Latex/images

where /home/shr/TableBank/data/Sampled_Detection_data/Latex/images is the input directory and jpg is the input format.

The detected table will be visualized in the directory /tmp/detectron-tablebank. By default, if no object can be detected for an input image, there will not be corresponding output file in the target directory. You can add ----always-out to generate the annotated pdf even if nothing is detected.