Skip to content

Latest commit

 

History

History
102 lines (70 loc) · 3.49 KB

README.md

File metadata and controls

102 lines (70 loc) · 3.49 KB

Official PyTorch implementation of "SequencePAR: Understanding Pedestrian Attributes via A Sequence Generation Paradigm"

SequencePAR: Understanding Pedestrian Attributes via A Sequence Generation Paradigm, Jiandong Jin, Xiao Wang *, Chenglong Li, Lili Huang, and Jin Tang

News:

Usage

Requirements

we use single RTX A100 40G GPU for training and evaluation.

Python 3.9.16
pytorch 1.12.1
torchvision 0.13.1
scipy 1.10.0
Pillow
easydict
torchtext

Dataset Preparation

Download the PETA dataset from here, PA100k dataset from here and RAP1 and RAP2 dataset form here, and We provide the processed WIDER dataset in here

Organize them in your dataset root dir folder as follows:

|-- your dataset root dir/
|   |-- <PETA>/
|       |-- images
|            |-- 00001.png
|            |-- 00002.png
|            |-- ...
|       |-- PETA.mat
|       |-- dataset_zs_run0.pkl
|
|   |-- <PA100k>/
|       |-- data
|            |-- 000001.jpg
|            |-- 000002.jpg
|            |-- ...
|       |-- annotation.mat
|
|   |-- <RAP1>/
|       |-- RAP_datasets
|       |-- RAP_annotation
|            |-- RAP_annotation.mat
|   |-- <RAP2>/
|       |-- RAP_datasets
|       |-- RAP_annotation
|            |-- RAP_annotation.mat
|       |-- dataset_zs_run0.pkl

Data Preparation

Run dataset/preprocess/peta_pad.py to get the dataset pkl file

python dataset/preprocess/peta_pad.py

We fill the images in the original dataset as a square with a simple black border fill and store it in Pad_datasets, you can read the original dataset directly and use the fill code we provided in AttrDataset.py. We provide processing code for the currently available publicly available pedestrian attribute identification dataset

Training

python train.py PETA

Evaluation

python eval.py PETA --check_point --dir your_dir

Abstract

Current pedestrian attribute recognition (PAR) algorithms are developed based on multi-label or multi-task learning frameworks, which aim to discriminate the attributes using specific classification heads. However, these discriminative models are easily influenced by imbalanced data or noisy samples. Inspired by the success of generative models, we rethink the pedestrian attribute recognition scheme and believe the generative models may perform better on modeling dependencies and complexity between human attributes. In this paper, we propose a novel sequence generation paradigm for pedestrian attribute recognition, termed SequencePAR. It extracts the pedestrian features using a pre-trained CLIP model and embeds the attribute set into query tokens under the guidance of text prompts. Then, a Transformer decoder is proposed to generate the human attributes by incorporating the visual features and attribute query tokens. The masked multi-head attention layer is introduced into the decoder module to prevent the model from remembering the next attribute while making attribute predictions during training. Extensive experiments on multiple widely used pedestrian attribute recognition datasets fully validated the effectiveness of our proposed SequencePAR.

Environment

Our Proposed Approach

Dataset

Experimental Results