tabnet

基于mmdetection的表格检测代码

install

Method 1

```shell conda create -n tabnet python=3.7 -y conda activate tabnet conda install pytorch torchvision torchaudio cudatoolkit=10.2 -c pytorch

pip install openmim mim install mmdet

```

Method 2

```shell conda create -n tabnetdev python=3.7 -y conda activate tabnetdev

conda install pytorch==1.6.0 torchvision==0.7.0 cudatoolkit=10.1 -c pytorch -y

pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu101/torch1.6.0/index.html

conda install pytorch==1.9.0 torchvision==0.10.0 cudatoolkit=11.1 -c pytorch -c nvidia -y pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu111/torch1.9.0/index.html

pip install -r requirements/build.txt

pip install -v -e . # or "python setup.py develop"

```

准备数据集

挂载群辉网络数据

```shell sudo mount.cifs //v4.vansin.top/datasets /datasets -o user=vansin,pass=*,vers=2.0 sudo mount.cifs //192.168.4.21/datasets /datasets -o user=vansin,pass=*,vers=2.0

sudo mount -t cifs -o username=vansin,password=Tml768300.,uid=$(id -u),gid=$(id -g) //v4.vansin.top/vansin /home/tml/datasets sudo mount -t cifs -o username=vansin,password=Tml768300.,uid=$(id -u),gid=$(id -g) //192.168.4.21/vansin /home/tml/datasets

mkdir /tmp/vansin

sudo mount -t cifs -o username=vansin,password=Tml768300.,uid=$(id -u),gid=$(id -g) //192.168.4.21/vansin /tmp/vansin

sudo mount -t cifs -o username=vansin,password=Tml768300.,uid=$(id -u),gid=$(id -g) //v4.vansin.top/vansin /tmp/vansin ln -s /tmp/vansin/datasets $(pwd)/data ln -s /tmp/vansin/workdirs $(pwd)/workdirs ```

软链接

shell ln -s /datasets/table0901/TableBank/Detection/ /home/tml/vansin/tabnet/data/table

Train

shell python tools/train.py configs/tabnet/table_v2.py

Test

Test 1 / 100 datasets

shell CUDA_VISIBLE_DEVICES=1 python test.py configs/tabnet/table_v4-mask_rcnn_swin-t-p4-w7_fpn_1x_coco_small.py \ /home/tml/vansin/paper/tabnet/work_dirs/table_v4-mask_rcnn_swin-t-p4-w7_fpn_1x_coco_small/epoch_12.pth \ --out results.pkl \ --eval bbox

Test all the train

shell CUDA_VISIBLE_DEVICES=1 python test.py configs/tabnet/table_v4-mask_rcnn_swin-t-p4-w7_fpn_1x_coco.py \ /home/tml/vansin/paper/tabnet/work_dirs/table_v4-mask_rcnn_swin-t-p4-w7_fpn_1x_coco/epoch_12.pth \ --out results.pkl \ --eval bbox

shell python demo/image_demo.py /datasets/table0901/TableBank/Detection/images/%5BMS-DOM2S%5D-180323_6.jpg \ work_dirs/faster_rcnn_r50_fpn_2x_coco/table_v3.py \ work_dirs/faster_rcnn_r50_fpn_2x_coco/epoch_2.pth \ --device cpu

```shell ln -s /tmp/ramdisk/Detection/ /home/tml/vansin/paper/tabnet/data/table

ln -s /run/user/1000/gvfs/smb-share:server=192.168.4.21,share=datasets/table0901/TableBank/Detection /home/tml/vansin/paper/tabnet/data/table ```

ONNX模型转换

Shell python tools/deployment/pytorch2onnx.py \ configs/tabnet/table_v4-mask_rcnn_swin-t-p4-w7_fpn_1x_3000.py \ checkpoints/tabnet/epoch_12.pth \ --output-file checkpoints/tabnet/epoch_12.onnx \ --input-img demo/demo.jpg \ --test-img tests/data/color.jpg \ --shape 608 608 \ --show \ --verify \ --dynamic-export \ --cfg-options \ model.test_cfg.deploy_nms_pre=-1

加速技巧

RAM虚拟硬盘

```shell

mkdir /tmp/vansinram sudo mount -t tmpfs -o size=72G tmpfs /tmp/vansinram ln -s /tmp/vansin_ram/datasets $(pwd)/data

```

结果分析技巧

结果分析

shell python tools/analysis_tools/analyze_results.py \ configs/tabnet/table_v7-mask_rcnn_swin-t-p4-w7_fpn_fp16_ms-crop-3x_icdar2013.py \ work_dirs/table_v7-mask_rcnn_swin-t-p4-w7_fpn_fp16_ms-crop-3x_icdar2013/epoch_34.pkl \ results

网络图可视化

shell python tools/deployment/pytorch2onnx.py \ ${CONFIG_FILE} \ ${CHECKPOINT_FILE} \ --output-file ${OUTPUT_FILE} \ --input-img ${INPUT_IMAGE_PATH} \ --shape ${IMAGE_SHAPE} \ --test-img ${TEST_IMAGE_PATH} \ --opset-version ${OPSET_VERSION} \ --cfg-options ${CFG_OPTIONS} --dynamic-export \ --show \ --verify \ --simplify