personreid sequential rl

An independent attempt to replicate the paper "Multi-shot Pedestrian Re-identification via Sequential Decision Making (CVPR2018)"

Network architecture

Person ReID as a Decision making problem

Prerequisites

torchvision
torch >= 0.4.0
visdom
tqdm
more-itertools

Data preparation

To prepare data under './data' folder, refer DATASET preparation from deep-person-reid repository.

MARS [8]:

1. Create a directory named mars/ under data/.
2. Download dataset to data/mars/ from http://www.liangzheng.com.cn/Project/project_mars.html.
3. Extract bbox_train.zip and bbox_test.zip.
4. Download split information from https://github.com/liangzheng06/MARS-evaluation/tree/master/info and put info/ in data/mars (we want to follow the standard split in [8]). The data structure would look like:

mars/
    bbox_test/
    bbox_train/
    info/

Use -d mars when running the training code.

iLIDS-VID [11]:

The code supports automatic download and formatting. Simple use -d ilidsvid when running the training code. The data structure would look like:

ilids-vid/
    i-LIDS-VID/
    train-test people splits/
    splits.json

PRID [12]:

1. Under data/, do mkdir prid2011 to create a directory.
2. Download dataset from https://www.tugraz.at/institute/icg/research/team-bischof/lrs/downloads/PRID11/ and extract it under data/prid2011.
3. Download the split created by iLIDS-VID from here, and put it in data/prid2011/. We follow [11] and use 178 persons whose sequences are more than a threshold so that results on this dataset can be fairly compared with other approaches. The data structure would look like:

prid2011/
    splits_prid2011.json
    prid_2011/
        multi_shot/
        single_shot/
        readme.txt

Use -d prid when running the training code.

DukeMTMC-VideoReID [16, 23]:

1. Make a directory data/dukemtmc-vidreid.
2. Download dukemtmc_videoReID.zip from https://github.com/Yu-Wu/DukeMTMC-VideoReID. Unzip the file to data/dukemtmc-vidreid.

You need to have

dukemtmc-vidreid/
    dukemtmc_videoReID/
        train_split/
        query_split/
        gallery_split/
        ... (and two license files)

Use -d dukemtmcvidreid when running the training code.

Stage-1 : Pre-training

Before posing the person Re-ID task as decision making problem, we need to be able to extract meaningful features from person images. Following transfer learning paradigm, in the pretraining stage, the imagenet pretrained AlexNet is finetuned using the particular video Re-ID dataset identities with the aid of person-classification cross-entropy loss and Triplet loss.

Take a look at the file src/doall_stage1_pretrain.py and the command-line options available in API stage1_parse_cmd_options() from src/cmd_options_parser.py to know more about stage1 training.

Stage-2 : DQN training

In stage-2, the Re-ID task is posed as a decision making problem and a Deep-Q Network is trained to maximize the Q-values of actions (SAME, DIFFERENT, UNSURE) with a reward function that encourages seeing less number of frames for the decision.

Take a look at the file src/doall_stage2_rl_training.py and the command-line options available in API stage2_parse_cmd_options() from src/cmd_options_parser.py to know more about stage2 training.

Note

This project code is trained and tested only for the dataset PRID2011, as training takes huge amount of time for large datasets (MARS, DukeMTMC-VideoReID) than I could afford at that time. Have a look at the project report and presentation for more details.

Credits

A large portion of dataloading, sampling codes are taken from the following repositories:

• deep-person-reid
• Video-Person-ReID