Timesformer Video Action Recognition Model - Open source and free, accurately recognize diverse actions in videos!

Timesformer Base Finetuned K400 Continual Lora Ucf101 Continual Lora Ucf101

Developed by NiiCole

A video action recognition model based on TimeSformer architecture, pre-trained on Kinetics-400 dataset and fine-tuned on UCF101 dataset

Downloads 18

Release Time : 10/12/2023

Model Overview

This model adopts the TimeSformer architecture, focusing on video action recognition tasks, with continual learning fine-tuning via LoRA technology.

Continual Learning Capability

Utilizes LoRA technology for continual learning, enabling adaptation to new tasks without forgetting previous knowledge

Efficient Video Processing

Based on TimeSformer architecture, specifically optimized for temporal feature extraction in videos

Transfer Learning Optimization

Pre-trained on the large-scale Kinetics-400 dataset and fine-tuned on UCF101

Video Action Classification

Temporal Feature Extraction

Cross-dataset Transfer Learning

Video Analysis

Action Recognition

Identify human actions and behaviors in videos

Achieves 96.99% accuracy on UCF101 validation set

Video Content Understanding

Analyze video content and extract key action information

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.7573	0.5	1138	0.6171	0.9349
0.2567	1.5	2276	0.3121	0.9699

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base