Contrastive Learning with Non-Semantic Negatives

This repository is the official implementation of Robust Contrastive Learning Using Negative Samples with Diminished Semantics. Contrastive learning utilizes positive pairs which preserve semantic information while perturbing superficial features in the training images. Similarly, we propose to generate negative samples to make the model more robust, where only the superfluous instead of the semantic features are preserved.

Preparation

Install PyTorch and check preprocess/ for ImageNet-100 and ImageNet-Texture preprocessing scripts.

Training

The following code is used to pre-train MoCo-v2 + patch / texture-based NS. The major code is developed with minimal modifications from the official implementation.

python moco-non-sem-neg.py -a resnet50 --lr 0.03 --batch-size 128 --dist-url 'tcp://localhost:10001' \
  --multiprocessing-distributed --world-size 1 --rank 0 \
  --mlp --moco-t 0.2 --aug-plus --cos --moco-k 16384 \
  --robust nonsem --num-nonsem 1 --alpha 2 --epochs 200 --patch-ratio 16 72 \
  --ckpt_path ./ckpts/mocov2_mocok16384_bs128_lr0.03_nonsem_16_72_noaug_nn1_alpha2_epoch200  \
  /path/to/imagenet-100/ 

python moco-non-sem-neg.py -a resnet50 --lr 0.03 --batch-size 128 --dist-url 'tcp://localhost:10001' \
  --multiprocessing-distributed --world-size 1 --rank 0 \
  --mlp --moco-t 0.2 --aug-plus --cos --moco-k 16384 \
  --robust texture_syn --num-nonsem 1 --alpha 2 --epochs 200 \
  --ckpt_path ./ckpts/mocov2_mocok16384_bs128_lr0.03_texture_nn1_alpha2_epoch200 \
  /path/to/imagenet-100-texture/

Change /path/to/imagenet-100/ with the ImageNet-100 dataset directory.
Change --alpha and -moco-k to reproduce results with different configurations.

Linear Evaluation

Run following code is used to reproduce MoCo-v2 + patch-based NS model reported in Table 1.

python main_lincls.py -a resnet50 --lr 10.0 --batch-size 128 --epochs 60 \
  --pretrained ./ckpts/mocov2_mocok16384_bs128_lr0.03_nonsem_16_72_noaug_nn1_alpha2_epoch200/checkpoint_0199.pth.tar \
  --dist-url 'tcp://localhost:10001' --multiprocessing-distributed --world-size 1 --rank 0 \
  --ckpt_path ./ckpts/mocov2_mocok16384_bs128_lr0.03_nonsem_16_72_noaug_nn1_alpha2_epoch200 \
  /path/to/imagenet-100/

Pre-trained Models

You can download pretrained models here:

	moco-k	alpha	ImageNet-100	Corruption	Sketch	Stylized	Rendition	Checkpoints
MoCo-v2	16384	-	77.88±0.28	43.08±0.27	28.24±0.58	16.20±0.55	32.92±0.12	Run1, Run2, Run3
+ Texture	16384	2	77.76±0.17	43.58±0.33	29.11±0.39	16.59±0.17	33.36±0.15	Run1, Run2, Run3
+ Patch	16384	2	79.35±0.12	45.13±0.35	31.76±0.88	17.37±0.19	34.78±0.15	Run1, Run2, Run3
+ Patch	16384	3	75.58±0.52	44.45±0.15	34.03±0.58	18.60±0.26	36.89±0.11	Run1, Run2, Run3
MoCo-v2	8192	-	77.73±0.38	43.22±0.39	28.45±0.36	16.83±0.12	33.19±0.44	Run1, Run2, Run3
+ Patch	8192	2	79.54±0.32	45.48±0.20	33.36±0.45	17.81±0.32	36.31±0.37	Run1, Run2, Run3

Contrastive Learning with Non-Semantic Negatives

Related tags

Overview

Contrastive Learning with Non-Semantic Negatives

Preparation

Training

Linear Evaluation

Pre-trained Models

Owner

CLIPImageClassifier wraps clip image model from transformers

Supplementary code for TISMIR paper "Sliding-Window Pitch-Class Histograms as a Means of Modeling Musical Form"

Another pytorch implementation of FCN (Fully Convolutional Networks)

Deep Sketch-guided Cartoon Video Inbetweening

Video Matting Refinement For Python

Molecular Sets (MOSES): A Benchmarking Platform for Molecular Generation Models

Code for "Localization with Sampling-Argmax", NeurIPS 2021

SSD: Single Shot MultiBox Detector pytorch implementation focusing on simplicity

RetinaNet-PyTorch - A RetinaNet Pytorch Implementation on remote sensing images and has the similar mAP result with RetinaNet in MMdetection

Point detection through multi-instance deep heatmap regression for sutures in endoscopy

A Real-ESRGAN equipped Colab notebook for CLIP Guided Diffusion

1st-in-MICCAI2020-CPM - Combined Radiology and Pathology Classification

MILK: Machine Learning Toolkit

A library for uncertainty representation and training in neural networks.

deep learning for image processing including classification and object-detection etc.

Implementation for "Manga Filling Style Conversion with Screentone Variational Autoencoder" (SIGGRAPH ASIA 2020 issue)

Image transformations designed for Scene Text Recognition (STR) data augmentation. Published at ICCV 2021 Workshop on Interactive Labeling and Data Augmentation for Vision.

A PyTorch implementation of unsupervised SimCSE

A CNN implementation using only numpy. Supports multidimensional images, stride, etc.

Writeups for the challenges from DownUnderCTF 2021