Skip to content

Latest commit

 

History

History
111 lines (91 loc) · 3.37 KB

README_NPU.md

File metadata and controls

111 lines (91 loc) · 3.37 KB
Raw

NPU doc

Training Dataset

COCO dataset

wget http://images.cocodataset.org/annotations/annotations_trainval2017.zip
wget http://images.cocodataset.org/zips/train2017.zip
wget http://images.cocodataset.org/annotations/image_info_test2017.zip
wget http://images.cocodataset.org/zips/val2017.zip
wget http://images.cocodataset.org/zips/test2017.zip

Extra files person_keypoints_256x192_resnet50_val2017_results.json

https://drive.google.com/drive/folders/10d8oSlCnWD-n3CBARXFj7kDBOKXfSYmf

result.json

https://drive.google.com/drive/folders/1blrmQ2CRiCeJDjx7sFzLFOBtlALf2Uxa

Model files

wget http://download.tensorflow.org/models/resnet_v1_50_2016_08_28.tar.gz
wget http://download.tensorflow.org/models/resnet_v1_101_2016_08_28.tar.gz
wget http://download.tensorflow.org/models/resnet_v1_152_2016_08_28.tar.gz

Repo

Follow the original repo guide to put all the data as the following structure. The person_keypoints_256x192_resnet50_val2017_results should replace the name_of_input_pose.json under ${POSE_ROOT}data/COCO/input_pose. (For the folder structure, may refer to https://github.com/mks0601/PoseFix_RELEASE for more details)

${POSE_ROOT}
|-- data
|-- |-- MPII
|   `-- |-- input_pose
|       |   |-- name_of_input_pose.json
|       |   |-- test_on_trainset
|       |   |   | -- result.json
|       |-- annotations
|       |   |-- train.json
|       |   `-- test.json
|       `-- images
|           |-- 000001163.jpg
|           |-- 000003072.jpg
|-- |-- PoseTrack
|   `-- |-- input_pose
|       |   |-- name_of_input_pose.json
|       |   |-- test_on_trainset
|       |   |   | -- result.json
|       |-- annotations
|       |   |-- train2018.json
|       |   |-- val2018.json
|       |   `-- test2018.json
|       |-- original_annotations
|       |   |-- train/
|       |   |-- val/
|       |   `-- test/
|       `-- images
|           |-- train/
|           |-- val/
|           `-- test/
|-- |-- COCO
|   `-- |-- input_pose
|       |   |-- name_of_input_pose.json
|       |   |-- test_on_trainset
|       |   |   | -- result.json
|       |-- annotations
|       |   |-- person_keypoints_train2017.json
|       |   |-- person_keypoints_val2017.json
|       |   `-- image_info_test-dev2017.json
|       `-- images
|           |-- train2017/
|           |-- val2017/
|           `-- test2017/
`-- |-- imagenet_weights
|       |-- resnet_v1_50.ckpt
|       |-- resnet_v1_101.ckpt
|       `-- resnet_v1_152.ckpt

NPU training command

cd scripts
sh run_npu_1p.sh

pbtxt and ckpt

checkpoint and pbtxt in https://drive.google.com/drive/folders/15ANznISxzazKQDdl2oz6jUqWLcBLodZz?usp=sharing

Finetune on your own dataset

You need to follow the following steps to train on your own dataset

  1. Create a folder in the data with the name your dataset and follow similar structure of COCO
  2. create a dataset.py in that folder.
  3. Change the config file, especially the paths for the data
  4. run the training command

evaluation result

Experiment AP AP(0.5) AP(0.75) APM APL AR AR(0.5) AR(0.75) ARM ARL
Original Paper 72.5 90.5 79.6 68.9 79.0 78.0 94.1 84.4 73.4 84.1
Trained model GPU 73.2 89.3 79.5 69.7 80.1 78.6 93.2 84.3 74.3 84.9
Trained model NPU 73.3 89.3 79.6 69.8 80.1 78.6 93.3 84.4 74.4 84.9