zero-shot-classification

An open source implementation of CLIP.

A collection of tutorials on state-of-the-art computer vision models and techniques. Explore everything from foundational architectures like ResNet to cutting-edge models like YOLO11, RT-DETR, SAM 2, Florence-2, PaliGemma 2, and Qwen2.5VL.

[ECCV2024] Video Foundation Models & Data for Multimodal Understanding

Diffusion Classifier leverages pretrained diffusion models to perform zero-shot classification without additional training

Cybertron: the home planet of the Transformers in Go

[NeurIPS 2023] This repository includes the official implementation of our paper "An Inverse Scaling Law for CLIP Training"

official code of “OpenShape: Scaling Up 3D Shape Representation Towards Open-World Understanding”

Reproducible scaling laws for contrastive language-image learning (https://arxiv.org/abs/2212.07143)

PyTorch code for MUST

[ICCV 2025] Official PyTorch Code for "Advancing Textual Prompt Learning with Anchored Attributes"

Multi-Aspect Vision Language Pretraining - CVPR2024

Unofficial (Golang) Go bindings for the Hugging Face Inference API

[ICASSP 2025] Open-source code for the paper "Enhancing Remote Sensing Vision-Language Models for Zero-Shot Scene Classification"

Evaluate custom and HuggingFace text-to-image/zero-shot-image-classification models like CLIP, SigLIP, DFN5B, and EVA-CLIP. Metrics include Zero-shot accuracy, Linear Probe, Image retrieval, and KNN accuracy.

[TPAMI 2023] Generative Multi-Label Zero-Shot Learning

Official PyTorch Implementation of MSDN (CVPR'22)

Low-latency ONNX and TensorRT based zero-shot classification and detection with contrastive language-image pre-training based prompts

Implementation of Z-BERT-A: a zero-shot pipeline for unknown intent detection.

[CVPR 2024] The official implementation of paper "Sculpting Holistic 3D Representation in Contrastive Language-Image-3D Pre-training"

Alternate Implementation for Zero Shot Text Classification: Instead of reframing NLI/XNLI, this reframes the text backbone of CLIP models to do ZSC. Hence, can be lightweight + supports more languages without trading-off accuracy. (Super simple, a 10th-grader could totally write this but since no 10th-grader did, I did) - Prithivi Da