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

UniFormer - official implementation of UniFormer

Official Implementation of "DialogLM: Pre-trained Model for Long Dialogue Understanding and Summarization."

Code for "Diffusion is All You Need for Learning on Surfaces"

Unimodal Face Classification with Multimodal Training

Code for the preprint "Well-classified Examples are Underestimated in Classification with Deep Neural Networks"

Listing arxiv - Personalized list of today's articles from ArXiv

New approach to benchmark VQA models

Uni-Fold: Training your own deep protein-folding models.

The MATH Dataset

torchsummaryDynamic: support real FLOPs calculation of dynamic network or user-custom PyTorch ops

DeepSpeed is a deep learning optimization library that makes distributed training easy, efficient, and effective.

Opinionated code formatter, just like Python's black code formatter but for Beancount

A new GCN model for Point Cloud Analyse

Face Mask Detection on Image and Video using tensorflow and keras

利用yolov5和TensorRT从0到1实现目标检测的模型训练到模型部署全过程

Simple and Robust Loss Design for Multi-Label Learning with Missing Labels

Rapid experimentation and scaling of deep learning models on molecular and crystal graphs.

LSTC: Boosting Atomic Action Detection with Long-Short-Term Context

Baselines for TrajNet++

Rafael Project- Classifying rockets to different types using data science algorithms.