kilianyp/triplet-reid-pytorch

A pytorch implementation of the "In Defense of the Triplet Loss for Person Re-Identification" paper (https://arxiv.org/abs/1703.07737). It also contains a implementation of the MGN network from the paper "Learning Discriminative Features with Multiple Granularities for Person Re-Identification". Reaches 83.17% mAP with MGN.

Description

A pytorch implementation of the "In Defense of the Triplet Loss for Person Re-Identification" paper (https://arxiv.org/abs/1703.07737).

This repository also contains also a implementation of the MGN network from the paper: "Learning Discriminative Features with Multiple Granularities for Person Re-Identification"

It is built from scratch with ease in experimenting, modification and reproducability. However, it still work in progress.

Requirements

• numpy
• Pillow
• h5py
• scipy
• torch
• torchvision

This repository depends for evaluation on the original repository: https://github.com/VisualComputingInstitute/triplet-reid

For evaluation:

• tensorflow
• scikit-learn
• scipy
• h5py

Installation

git clone https://github.com/kilsenp/triplet-reid-pytorch.git pip install -r requirements.txt (for training)

Train

python3 train.py --csv_file path/to/ 
                 --data_dir path/to/image/base/directory 
                 --loss BatchHard
                 --model trinet

The script looks for the image files by joining the data_dir and image path given in the csv file. The csv file should be a two column file with pids, fids.

For market, you can find them here:

If you would like to train using the MGN network, use the following command:

python3 train.py --csv_file path/to/ 
                 --data_dir path/to/image/base/directory 
                 --loss BatchHardSingleWithSoftmax
                 --model mgn
                 --mgn branches 1 2 3
                 --dim 256

Evaluation

You can use embed.py to write out embeddings that are compatible with the evaluation script.

python3 embed.py --csv_file path/to/gallery/or/query/csv
                 --data_dir path/to/image/base/directory
                 --model path/to/model/file

To calculate the final scores, please use the evaluation script from here!

Scores without Re-rank (and pretrained models)

Market-1501

Trinet

Settings:

• P=18
• K=4
• dim=128

Download Model (GoogleDrive)

Test time augmentation	mAP	top-1	top-5	top-10
None	65.06%	80.31%	92.25%	94.71%
With TenCrop	69.44%	83.40%	93.59%	96.17%

MGN

Settings:

  "K": 4,
  "P": 16,
  "checkpoint_frequency": 1000,
  "commit": "c1c1a27",
  "csv_file": "data/market1501_train.csv",
  "data_dir": "datasets/Market-1501",
  "decay_start_iteration": 15000,
  "dim": 256,
  "experiment": "new_mgn_123_single",
  "image_height": 384,
  "image_width": 128,
  "log_level": 1,
  "loss": "BatchHardSingleWithSoftmax",
  "lr": 0.0003,
  "margin": "1.2",
  "mgn_branches": [
    1,
    2,
    3
  ],
  "model": "mgn",
  "no_multi_gpu": false,
  "num_classes": 751,
  "output_path": "training",
  "restore_checkpoint": null,
  "sampler": "TripletBatchSampler",
  "scale": 1.0,
  "train_iterations": 25000

Test time augmentation	mAP	top-1	top-5	top-10
With Horizontal Flip	83.17%	93.62%	97.86%	98.66%

TODO

• Evaluate current MGN
• Upload MGN model
• Improve logging (Use tensorboard or similar)
• Clean up CPU GPU mess