Object-aware Contrastive Learning for Debiased Scene Representation

Last update: Dec 14, 2022

Overview

Object-aware Contrastive Learning

Official PyTorch implementation of "Object-aware Contrastive Learning for Debiased Scene Representation" by Sangwoo Mo*, Hyunwoo Kang*, Kihyuk Sohn, Chun-Liang Li, and Jinwoo Shin.

Installation

Install required libraries.

pip install -r requirements.txt

Download datasets in /data (e.g., /data/COCO).

Train models

Logs will be saved in logs/{dataset}_{model}_{arch}_b{global_batch_size} directory, where global_batch_size = num_nodes * gpus * batch_size (default batch size = 64 * 4 = 256).

Step 1. Train vanilla models

Train vanilla models (change dataset and ft_datasets as cub or in9).

python pretrain.py --dataset coco --model moco --arch resnet18\
    --ft_datasets coco --batch_size 64 --max_epochs 800

Step 2. Pre-compute CAM masks

Pre-compute bounding boxes for object-aware random crop.

python inference.py --mode save_box --model moco --arch resnet18\
    --ckpt_name coco_moco_r18_b256 --dataset coco\
    --expand_res 2 --cam_iters 10 --apply_crf\
    --save_path data/boxes/coco_cam-r18.txt

Pre-compute masks for background mixup.

python inference.py --mode save_mask --model moco --arch resnet18\
    --ckpt_name in9_moco_r18_256 --dataset in9\
    --expand_res 1 --cam_iters 1\
    --save_path data/masks/in9_cam-r18

Step 3. Re-train debiased models

Train contextual debiased model with object-aware random crop.

python pretrain.py --dataset coco-box-cam-r18 --model moco --arch resnet18\
     --ft_datasets coco --batch_size 64 --max_epochs 800

Train background debiased model with background mixup.

python pretrain.py --dataset in9-mask-cam-r18 --model moco_bgmix --arch resnet18\
    --ft_datasets in9 --batch_size 64 --max_epochs 800

Evaluate models

Linear evaluation

python inference.py --mode lineval --model moco --arch resnet18\
    --ckpt_name coco_moco_r18_b256 --dataset coco

Object localization

python inference.py --mode seg --model moco --arch resnet18\
    --ckpt_name cub200_moco_r18_b256 --dataset cub200\
    --expand_res 2 --cam_iters 10 --apply_crf

Detection & Segmentation (fine-tuning)

mv detection
python convert-pretrain-to-detectron2.py coco_moco_r50.pth coco_moco_r50.pkl
python train_net.py --config-file configs/coco_R_50_C4_2x_moco.yaml --num-gpus 8\
    MODEL.WEIGHTS weights/coco_moco_r18.pkl

Object-aware Contrastive Learning for Debiased Scene Representation

Related tags

Overview

Object-aware Contrastive Learning

Installation

Train models

Step 1. Train vanilla models

Step 2. Pre-compute CAM masks

Step 3. Re-train debiased models

Evaluate models

Linear evaluation

Object localization

Detection & Segmentation (fine-tuning)

Owner

A fuzzing framework for SMT solvers

This repository provides code for "On Interaction Between Augmentations and Corruptions in Natural Corruption Robustness".

A data-driven approach to quantify the value of classifiers in a machine learning ensemble.

AutoML library for deep learning

[ICCV2021] IICNet: A Generic Framework for Reversible Image Conversion

Official code of our work, AVATAR: A Parallel Corpus for Java-Python Program Translation.

Official implementation of VQ-Diffusion

MinkLoc3D-SI: 3D LiDAR place recognition with sparse convolutions,spherical coordinates, and intensity

Solver for Large-Scale Rank-One Semidefinite Relaxations

This project aims to segment 4 common retinal lesions from Fundus Images.

Code for our paper 'Generalized Category Discovery'

Evaluation toolkit of the informative tracking benchmark comprising 9 scenarios, 180 diverse videos, and new challenges.

Self-Supervised Multi-Frame Monocular Scene Flow (CVPR 2021)

MetaBalance: High-Performance Neural Networks for Class-Imbalanced Data

Pytorch cuda extension of grid_sample1d

Implementation of ML models like Decision tree, Naive Bayes, Logistic Regression and many other

Implementation of Transformer in Transformer, pixel level attention paired with patch level attention for image classification, in Pytorch

Official repo of the paper "Surface Form Competition: Why the Highest Probability Answer Isn't Always Right"

Decision Transformer: A brand new Offline RL Pattern

Artstation-Artistic-face-HQ Dataset (AAHQ)