Update README.md

.gitignore

@@ -8,8 +8,9 @@ log/
 log
 logs/
 logs
+models/
 datasets/*
 **/*.c
 **/*.linux-x86_64*
 **/*.egg-info
-**/*.egg
+**/*.egg

projects/MGH/README.md

@@ -1,16 +1,22 @@
 # 项目
-CAP文件夹中包含**When Vision Meets Meta Information: Meta Hypergraph Guided Network for Unsupervised Re-identification**的具体实现，
+MGH文件夹中包含**MGH: Metadata Guided Hypergraph Modeling for Unsupervised Person Re-identification**的具体实现。
+
 该项目以AAAI2021文章**Camera-aware Proxies for Unsupervised Person Re-Identification**为基础，针对行人重识别的无监督领域自适应问题进行了如下修改：
 1. 使用超图对聚类标签进行了修正；
 2. 增加了基于memory的listwise loss，用于近似优化mAP。
+
 整体的pipeline为**特征提取->聚类->赋伪标签->训练**的循环执行。
 
 ## 说明
 
 ### 环境配置
+项目根目录为
+```
+ROOT_DIR
+```
 环境配置请参考项目根目录下
 ```
-docs/GETTING_STARTED.md
+$ROOT_DIR/docs/GETTING_STARTED.md
 ```
 建议使用conda进行项目管理，需要注意的是安装apex建议通过编译的方式，在CUDA版本较高的情况下apex目前没有提供安装包。
 在运行make.sh时，编译了Cython文件用于加速测试，也编译了GNN Re-ranking。
@@ -19,13 +25,35 @@ docs/GETTING_STARTED.md
 本项目在Market1501、DukeMTMC-reid、MSMT17上进行了验证。
 数据集配置请参考路径
 ```
-datasets/README.md
+$ROOT_DIR/datasets/README.md
 ```
 
-### 运行
-代码运行入口为projects/CAP/train_net.py
-训练与测试脚本分别为projects/CAP/train.sh和projects/CAP/test.sh，
-如需要进行修改，请参考以上两个文件。
+### 训练
+代码运行入口为projects/MGH/train_net.py
+训练脚本为projects/MGH/train.sh，
+执行训练脚本
+```bash
+cd $ROOT_DIR
+bash projects/MGH/reproduce.sh
+```
+
+### 测试
+测试脚本为projects/MGH/test.sh，
+执行测试脚本前请下载已经训练好的模型models.zip并解压至$ROOT_DIR/models，文件目录如下
+```
+.
+├── duke
+│   └── model_duke.pth
+├── market
+│   └── model_market.pth
+└── msmt17
+    └── model_msmt17.pth
+```
+运行测试脚本
+```bash
+cd $ROOT_DIR
+bash project/MGH/test.sh
+```
 
 ### 代码结构
 本项目仿照fastreid中其他项目，继承fastreid已有类及其函数。

projects/MGH/reproduce.sh

@@ -0,0 +1,7 @@
+#!/bin/bash
+# market
+CUDA_VISIBLE_DEVICES=0 python projects/CAP/train_net.py --num-gpus 1 --config-file projects/CAP/configs/Market1501/BoT_R50.yml OUTPUT_DIR logs/MGH/market1501 PSEUDO.NAME "dbscan_hg_lp" CAP.LOSS_IDENTITY.SCALE 1.0 CAP.LOSS_IDENTITY.START_EPOCH 5 CAP.LOSS_CAMERA.SCALE 1.0 CAP.LOSS_CAMERA.START_EPOCH 0 CAP.LOSS_CAMERA.WEIGHTED True CAP.LOSS_INSTANCE.SCALE 10.0 CAP.LOSS_INSTANCE.START_EPOCH 5 CAP.LOSS_INSTANCE.NAME "smoothaploss" CAP.LOSS_INSTANCE.LIST_LENGTH 1000
+# dukemtmc
+CUDA_VISIBLE_DEVICES=0 python projects/CAP/train_net.py --num-gpus 1 --config-file projects/CAP/configs/DukeMTMC/BoT_R50.yml OUTPUT_DIR logs/MGH/dukemtmc PSEUDO.NAME "dbscan_hg_lp" CAP.LOSS_IDENTITY.SCALE 1.0 CAP.LOSS_IDENTITY.START_EPOCH 5 CAP.LOSS_CAMERA.SCALE 1.0 CAP.LOSS_CAMERA.START_EPOCH 0 CAP.LOSS_CAMERA.WEIGHTED True CAP.LOSS_INSTANCE.SCALE 100.0 CAP.LOSS_INSTANCE.START_EPOCH 5 CAP.LOSS_INSTANCE.NAME "smoothaploss" CAP.LOSS_INSTANCE.LIST_LENGTH 1000
+# msmt
+CUDA_VISIBLE_DEVICES=0 python projects/CAP/train_net.py --num-gpus 1 --config-file projects/CAP/configs/MSMT17/BoT_R50.yml OUTPUT_DIR logs/MGH/msmt17 PSEUDO.NAME "dbscan_hg_lp" CAP.LOSS_IDENTITY.SCALE 1.0 CAP.LOSS_IDENTITY.START_EPOCH 5 CAP.LOSS_CAMERA.SCALE 1.0 CAP.LOSS_CAMERA.START_EPOCH 0 CAP.LOSS_CAMERA.WEIGHTED True CAP.LOSS_INSTANCE.SCALE 100.0 CAP.LOSS_INSTANCE.START_EPOCH 5 CAP.LOSS_INSTANCE.NAME "smoothaploss" CAP.LOSS_INSTANCE.LIST_LENGTH 1000

projects/MGH/test.sh

@@ -1,4 +1,4 @@
 # testing
-CUDA_VISIBLE_DEVICES=0 python projects/MGH/train_net.py --eval-only --num-gpus 1 --config-file projects/MGH/configs/Market1501/BoT_R50.yml OUTPUT_DIR logs/MGH/market1501_test MODEL.WEIGHTS logs/CAP/BoT_R50_new/market1501_hg_knn_camera_1.0identity5_weighted_1.0camera0_10.0instance_smoothaploss5/model_0018399.pth CAP.ST_TEST True
-CUDA_VISIBLE_DEVICES=1 python projects/MGH/train_net.py --eval-only --num-gpus 1 --config-file projects/MGH/configs/DukeMTMC/BoT_R50.yml OUTPUT_DIR logs/MGH/dukemtmc_test MODEL.WEIGHTS logs/CAP/BoT_R50/dukemtmc_hg_1.0identity5_weighted_1.0camera0_10.0instance_smoothaploss5/model_0019599.pth CAP.ST_TEST True
-CUDA_VISIBLE_DEVICES=2 python projects/MGH/train_net.py --eval-only --num-gpus 1 --config-file projects/MGH/configs/MSMT17/BoT_R50.yml OUTPUT_DIR logs/MGH/msmt17_test MODEL.WEIGHTS logs/CAP/BoT_R50_k1_30_w_cls/msmt17_hg_1.0identity5_weighted_1.0camera0_10.0instance_smoothaploss5/model_0019999.pth CAP.ST_TEST True
+CUDA_VISIBLE_DEVICES=0 python projects/MGH/train_net.py --eval-only --num-gpus 1 --config-file projects/MGH/configs/Market1501/BoT_R50.yml OUTPUT_DIR logs/MGH/market1501_test MODEL.WEIGHTS models/market/model_market.pth CAP.ST_TEST True
+CUDA_VISIBLE_DEVICES=1 python projects/MGH/train_net.py --eval-only --num-gpus 1 --config-file projects/MGH/configs/DukeMTMC/BoT_R50.yml OUTPUT_DIR logs/MGH/dukemtmc_test MODEL.WEIGHTS models/duke/model_duke.pth CAP.ST_TEST True
+CUDA_VISIBLE_DEVICES=2 python projects/MGH/train_net.py --eval-only --num-gpus 1 --config-file projects/MGH/configs/MSMT17/BoT_R50.yml OUTPUT_DIR logs/MGH/msmt17_test MODEL.WEIGHTS models/msmt/model_msmt.pth CAP.ST_TEST True