DTennant/reid_baseline_with_syncbn DTennant/reid_baseline_with_syncbn

  • This repository has been archived on 24/Apr/2023
  • Stars
    star
    157
  • Rank 238,399 (Top 5 %)
  • Language
    Python
  • License
    MIT License
  • Created over 5 years ago
  • Updated over 4 years ago
Twitter logo Share LinkedIn logo Share Facebook logo Share
DTennant/reid_baseline_with_syncbn
github

DTennant

Reviews

There are no reviews yet. Be the first to send feedback to the community and the maintainers!

Repository Details

Reimplementation of Bag of Tricks and A Strong Baseline for Deep Person Re-identification

A Modified reid_baseline supporting Multi-GPU and SyncBN training

Original repo here

However, the original repo uses ignite for training and saving the model, which is incompatible with model using SyncBN, So I reimplement the baseline without the use of ignite.(???)

Some code in this repo is borrowed from the original one.

Usage

  1. Clone the repo using git clone
  2. Compile the code for Cython accelerated evaluation code cd evaluate/eval_cylib && make
  3. the SyncBN module is pure pytorch implementation, so no need to compile once you have pytorch.
  4. Modify the training config in configs folder.
  5. Start training:
# training with only one GPU
CUDA_VISIBLE_DEVICES=1 python main.py -c configs/debug.yml
# testing with one GPU
CUDA_VISIBLE_DEVICES=1 python main.py -t -c configs/debug.yml TEST.WEIGHT /path/to/saved/weights

# training with multi-GPU
CUDA_VISIBLE_DEVICES=1,2 python main.py -c configs/debug_multi-gpu.yml
# testing with multi-GPU
CUDA_VISIBLE_DEVICES=1,2 python main.py -t -c configs/debug_multi-gpu.yml TEST.WEIGHT /path/to/saved/weights

# fp16 training
CUDA_VISIBLE_DEVICES=1 python main.py -c configs/debug.yml SOLVER.FP16 True

Result

I only trained the model with 30 epoch, so the model may not be fully converged.

Note that the Resnet50 model is trained with Warmup, Random erasing augmentation, Last stride=1 and BNNeck.

Result training with one GPU

2019-07-06 13:43:29,884 reid_baseline.eval INFO: Validation Result:
2019-07-06 13:43:29,885 reid_baseline.eval INFO: CMC Rank-1: 89.55%
2019-07-06 13:43:29,885 reid_baseline.eval INFO: CMC Rank-5: 96.76%
2019-07-06 13:43:29,885 reid_baseline.eval INFO: CMC Rank-10: 98.28%
2019-07-06 13:43:29,885 reid_baseline.eval INFO: mAP: 74.50%
2019-07-06 13:43:29,885 reid_baseline.eval INFO: --------------------
2019-07-06 13:45:40,783 reid_baseline.eval INFO: ReRanking Result:
2019-07-06 13:45:40,783 reid_baseline.eval INFO: CMC Rank-1: 92.67%
2019-07-06 13:45:40,783 reid_baseline.eval INFO: CMC Rank-5: 96.41%
2019-07-06 13:45:40,784 reid_baseline.eval INFO: CMC Rank-10: 97.48%
2019-07-06 13:45:40,784 reid_baseline.eval INFO: mAP: 90.00%
2019-07-06 13:45:40,784 reid_baseline.eval INFO: --------------------

Result training with multi-GPU

2019-07-06 14:24:32,171 reid_baseline.eval INFO: Use multi gpu to inference
2019-07-06 14:25:41,053 reid_baseline.eval INFO: Validation Result:
2019-07-06 14:25:41,054 reid_baseline.eval INFO: CMC Rank-1: 87.44%
2019-07-06 14:25:41,054 reid_baseline.eval INFO: CMC Rank-5: 96.17%
2019-07-06 14:25:41,054 reid_baseline.eval INFO: CMC Rank-10: 97.54%
2019-07-06 14:25:41,054 reid_baseline.eval INFO: mAP: 72.13%
2019-07-06 14:25:41,054 reid_baseline.eval INFO: --------------------
2019-07-06 14:27:23,449 reid_baseline.eval INFO: ReRanking Result:
2019-07-06 14:27:23,450 reid_baseline.eval INFO: CMC Rank-1: 91.30%
2019-07-06 14:27:23,450 reid_baseline.eval INFO: CMC Rank-5: 95.64%
2019-07-06 14:27:23,450 reid_baseline.eval INFO: CMC Rank-10: 96.79%
2019-07-06 14:27:23,450 reid_baseline.eval INFO: mAP: 88.92%
2019-07-06 14:27:23,450 reid_baseline.eval INFO: --------------------

Result training with FP16

2019-07-07 12:29:20,649 reid_baseline.eval INFO: Validation Result:
2019-07-07 12:29:20,650 reid_baseline.eval INFO: CMC Rank-1: 90.44%
2019-07-07 12:29:20,650 reid_baseline.eval INFO: CMC Rank-5: 96.47%
2019-07-07 12:29:20,650 reid_baseline.eval INFO: CMC Rank-10: 98.13%
2019-07-07 12:29:20,650 reid_baseline.eval INFO: mAP: 75.88%
2019-07-07 12:29:20,650 reid_baseline.eval INFO: --------------------
2019-07-07 12:31:01,269 reid_baseline.eval INFO: ReRanking Result:
2019-07-07 12:31:01,270 reid_baseline.eval INFO: CMC Rank-1: 93.14%
2019-07-07 12:31:01,270 reid_baseline.eval INFO: CMC Rank-5: 96.26%
2019-07-07 12:31:01,270 reid_baseline.eval INFO: CMC Rank-10: 97.12%
2019-07-07 12:31:01,270 reid_baseline.eval INFO: mAP: 90.22%
2019-07-07 12:31:01,270 reid_baseline.eval INFO: --------------------

Result training on Market1501, testing on Partial-iLIDS

2019-07-08 18:15:14,299 reid_baseline.eval INFO: Validation Result:
2019-07-08 18:15:14,299 reid_baseline.eval INFO: CMC Rank-1: 45.38%
2019-07-08 18:15:14,300 reid_baseline.eval INFO: CMC Rank-3: 57.14%
2019-07-08 18:15:14,300 reid_baseline.eval INFO: mAP: 55.76%
2019-07-08 18:15:14,300 reid_baseline.eval INFO: --------------------
2019-07-08 18:15:14,678 reid_baseline.eval INFO: ReRanking Result:
2019-07-08 18:15:14,678 reid_baseline.eval INFO: CMC Rank-1: 28.57%
2019-07-08 18:15:14,678 reid_baseline.eval INFO: CMC Rank-3: 48.74%
2019-07-08 18:15:14,678 reid_baseline.eval INFO: mAP: 42.62%
2019-07-08 18:15:14,679 reid_baseline.eval INFO: --------------------

Result on VeRI-WILD small test set

2019-07-12 22:47:57,250 reid_baseline.eval INFO: Validation Result:
2019-07-12 22:47:57,250 reid_baseline.eval INFO: CMC Rank-1: 90.43%
2019-07-12 22:47:57,250 reid_baseline.eval INFO: CMC Rank-5: 96.85%
2019-07-12 22:47:57,250 reid_baseline.eval INFO: CMC Rank-10: 98.19%
2019-07-12 22:47:57,250 reid_baseline.eval INFO: mAP: 74.27%
2019-07-12 22:47:57,251 reid_baseline.eval INFO: --------------------
2019-07-12 22:57:10,006 reid_baseline.eval INFO: ReRanking Result:
2019-07-12 22:57:10,006 reid_baseline.eval INFO: CMC Rank-1: 89.66%
2019-07-12 22:57:10,006 reid_baseline.eval INFO: CMC Rank-5: 95.45%
2019-07-12 22:57:10,006 reid_baseline.eval INFO: CMC Rank-10: 97.46%
2019-07-12 22:57:10,007 reid_baseline.eval INFO: mAP: 77.42%
2019-07-12 22:57:10,007 reid_baseline.eval INFO: --------------------

Result on VeRI776

2019-07-13 11:35:12,402 reid_baseline.eval INFO: Validation Result:
2019-07-13 11:35:12,402 reid_baseline.eval INFO: CMC Rank-1: 95.65%
2019-07-13 11:35:12,402 reid_baseline.eval INFO: CMC Rank-5: 97.91%
2019-07-13 11:35:12,403 reid_baseline.eval INFO: CMC Rank-10: 99.11%
2019-07-13 11:35:12,403 reid_baseline.eval INFO: mAP: 77.15%
2019-07-13 11:35:12,403 reid_baseline.eval INFO: --------------------
2019-07-13 11:36:02,050 reid_baseline.eval INFO: ReRanking Result:
2019-07-13 11:36:02,050 reid_baseline.eval INFO: CMC Rank-1: 97.20%
2019-07-13 11:36:02,050 reid_baseline.eval INFO: CMC Rank-5: 97.85%
2019-07-13 11:36:02,050 reid_baseline.eval INFO: CMC Rank-10: 98.75%
2019-07-13 11:36:02,050 reid_baseline.eval INFO: mAP: 81.74%
2019-07-13 11:36:02,050 reid_baseline.eval INFO: --------------------

More Repositories

1

CL-Visualizing-Feature-Transformation

Improving Contrastive Learning by Visualizing Feature Transformation, ICCV 2021 Oral
Jupyter Notebook
89
star
2

distill_visual_priors

2nd place solution of ECCV 2020 workshop VIPriors Image Classification Challenge, https://arxiv.org/abs/2008.00261
Python
13
star
3

tl-YOLOv2

A tensorlayer implementation of YOLOv2: Object Detection for both image and video!
Python
12
star
4

dual-rank-ncd

Novel Visual Category Discovery with Dual Ranking Statistics and Mutual Knowledge Distillation. Bingchen Zhao and Kai Han. (NeurIPS 2021)
Python
11
star
5

GPC

Python
10
star
6

Incremental-Generalized-Category-Discovery

Python
8
star
7

dt-tradeoff

[ECCV 2022] Discriminability-Transferability Trade-Off: An Information-Theoretic Perspective
Python
6
star
8

gpu-docker

nvidia-docker image to avoid some troublesome deep learning setting up and incompatible issues
Dockerfile
4
star
9

fpga_final

Verilog
3
star
10

GPT4-V-selenium

Python
3
star
11

CycleGAN

A implementation of CycleGAN(https://arxiv.org/abs/1703.10593)
Python
1
star
12

MIRB_eval

Python
1
star