Data

Tools

Indexes

Applications

Experiments

Open Source Science

https://github.com/1btu/pruned_lightweight_openpose

The pruned version of lightweight OpenPose.

https://github.com/1btu/pruned_lightweight_openpose

Science Score: 10.0%

This score indicates how likely this project is to be science-related based on various indicators:

  • ○
    CITATION.cff file
  • ○
    codemeta.json file
  • ○
    .zenodo.json file
  • ○
    DOI references
  • ✓
    Academic publication links
    Links to: arxiv.org
  • ○
    Academic email domains
  • ○
    Institutional organization owner
  • ○
    JOSS paper metadata
  • ○
    Scientific vocabulary similarity
    Low similarity (9.5%) to scientific vocabulary
Last synced: 10 months ago · JSON representation

Repository

The pruned version of lightweight OpenPose.

Basic Info
  • Host: GitHub
  • Owner: 1BTU
  • Default Branch: master
  • Size: 152 MB
Statistics
  • Stars: 0
  • Watchers: 0
  • Forks: 0
  • Open Issues: 0
  • Releases: 0
Fork of ChrisAllenMing/pruned_lightweight_openpose
Created about 3 years ago · Last pushed about 7 years ago

https://github.com/1BTU/pruned_lightweight_openpose/blob/master/ 

# A Pruned Version for Lightweight OpenPose

This repository contains channel-pruned models for lightweight OpenPose ([Real-time 2D Multi-Person Pose Estimation on CPU: Lightweight OpenPose](https://arxiv.org/pdf/1811.12004.pdf)), and it mainly follows the work of [Daniil Osokin](https://github.com/Daniil-Osokin/lightweight-human-pose-estimation.pytorch). The channel pruning code is based on our ICCV 2019 submission which will be open-source after acceptance. 



  



## Table of Contents

* [Requirements](#requirements)
* [Prerequisites](#prerequisites)
* [Training](#training)
* [Validation](#validation)
* [Demo](#demo)
* [Pruned model](#pruned_model)
* [Fine-tuned model](#fine-tuned_model)
* [Unpruned pre-trained model](#pre-trained-model)

## Requirements

* Ubuntu 16.04
* Python 3.6
* PyTorch 0.4.1 (should also work with 1.0, but not tested)

## Prerequisites

1. Download COCO 2017 dataset: [http://cocodataset.org/#download](http://cocodataset.org/#download) (train, val, annotations) and unpack it to `` folder.
2. Install requirements `pip install -r requirements.txt`

## Training

1. Fine-tune the pruned model. Run `CUDA_VISIBLE_DEVICES= python train_prune.py --train-images-folder ./coco/train2017/ --prepared-train-labels prepared_train_annotation.pkl --val-labels val_subset.json --val-images-folder ./coco/val2017/ --checkpoint-path ./pruned_models/ --num-refinement-stages 3 --experiment-name  --weights-only`

## Validation

1. For training-time synchronous validation. Run `CUDA_VISIBLE_DEVICES= python val_per_epoch.py`
2. Validation for a specific checkpoint. Run `python val_prune_oneepoch.py --labels /annotations/person_keypoints_val2017.json --images-folder /val2017 --checkpoint-path `

## Demo
1. For a simple demo. Run `python demo.py --checkpoint-path ./fine-tuned_models/ --images `

## Pruned model 

We provide two pruned models with different compression rate: ./pruned_models/0.3.pth.tar (reduce 15.92% flops) and ./pruned_models/0.8.tar.pth (reduce 25.6% flops).

## Fine-tuned model 

The model fine-tuned from the pruned model `./pruned_models/0.3.pth.tar` is available in `./fine-tuned_models/`. 

## Unpruned pre-trained model 

The model expects normalized image (mean=[128, 128, 128], scale=[1/256, 1/256, 1/256]) in planar BGR format.
Pre-trained on COCO model is available at: ./pre-trained_models/checkpoint_iter_370000.pth.tar, it has 40% of AP on COCO validation set (38.6% of AP on the val *subset*).

Owner

  • Name: leonming
  • Login: 1BTU
  • Kind: user

GitHub Events

Total
Last Year