timesformer

Transformers 3rd Edition

PyTorch implementation of a collections of scalable Video Transformer Benchmarks.

TianChi AIEarth Contest Solution

Code for Video Deepfake Detector from "MINTIME: Multi-Identity Size-Invariant Video Deepfake Detection", paper available on IEEE Transactions on Information Forensics and Security.

The notebook explains the various steps to obtain the results of publication: "Is Space-Time Attention All You Need for Video Understanding?"

Extended implementation of the Vesuvius Challenge 2023 Grand Prize winner. This project features extra functionality compared to the original ink detection script, giving users fine control over the layers they perform inference on.

Experimental fork of TimeSformer from Facebook AI to extend the attention-based model to video generation.

A research implementation of a context-conditioned, zero-shot video anomaly detection framework that integrates spatiotemporal features extracted via TimeSformer with contrastive predictive coding and semantic alignment through CLIP. The repository includes training and evaluation pipelines, configuration files, and the accompanying thesis/paper.

This app allows you to upload a video, converts it into frames, and predicts the action using a pre-trained model. We use TimeSformer, a state-of-the-art video transformer model, which processes video frames as a sequence of images and captures temporal relationships to predict actions effectively. Experience seamless action recognition with visual

time-series-forecast-chronos

Action recognition ツール類

Experimental fork of TimeSformer from Facebook AI to extend the attention-based model to video generation.

Trained a TimeSformer to classify 8 cricket shot types from 4,200 clips with MediaPipe pose for action-boundary detection; deployed via FastAPI + React

Reverse video search using TimeSformer transformer embeddings and FAISS vector indexing. Upload a video, retrieve visually similar clips from UCF-101 in milliseconds.

Deep learning models for 3d volumetric ink detection on ancient Vesuvius scroll fragments.

Action recognition using VideoMAE, TimeSFormer and ViViT | 1,250 video clips | 93.62% Top-1 | 98.40% Top-5 | spatio-temporal localisation

This app allows you to upload a video, converts it into frames, and predicts the action using a pre-trained model. We use TimeSformer, a state-of-the-art video transformer model, which processes video frames as a sequence of images and captures temporal relationships to predict actions effectively. Experience seamless action recognition with visual

SceneSolver is an AI-powered surveillance intelligence platform designed to assist investigators and security personnel in analyzing surveillance footage. Building on initial exploration and early prototype work, the project evolved into an end-to-end multimodal pipeline that combines video anomaly detection, temporal activity recognition, object d