Vivit-b-16x2-Kinetics400-UCF-Crime Open-Source Model - Free for Video Anomaly Detection

Vivit B 16x2 Kinetics400 UCF Crime

Developed by prathameshdalal

This model is a fine-tuned version of google/vivit-b-16x2-kinetics400 on the UCF-Crime dataset for video anomaly detection tasks.

Downloads 506

Release Time : 9/22/2023

Model Overview

A video classification model based on the ViViT architecture, specifically fine-tuned for anomaly behavior detection in the UCF-Crime dataset.

Video Anomaly Detection

Specifically designed to detect abnormal behaviors in surveillance videos

Transformer-based Architecture

Adopts the ViViT architecture, leveraging the advantages of Transformers in video understanding

Pre-trained Fine-tuning

Based on the Kinetics-400 pre-trained model, fine-tuned on the UCF-Crime dataset

Video Classification

Anomaly Behavior Detection

Surveillance Video Analysis

Public Safety

Surveillance Video Anomaly Detection

Automatically detects abnormal behaviors in surveillance videos, such as violent incidents, theft, etc.

Achieves 61.49% accuracy on the UCF-Crime evaluation set

Training Loss	Epoch	Step	Validation Loss	Accuracy
1.2072	0.06	196	1.6400	0.5518
1.5513	1.06	392	1.4988	0.5634
1.1038	2.06	588	1.5328	0.5861
0.9462	3.06	784	1.3932	0.6178
0.7387	4.06	980	1.5449	0.6060
0.5085	5.06	1176	1.3075	0.6287
0.4443	6.06	1372	1.6743	0.6001
0.4695	7.06	1568	1.5287	0.6172
0.4409	8.06	1764	1.7749	0.6089
0.1158	9.06	1960	1.9027	0.6076
0.1183	10.06	2156	1.9622	0.6085
0.1322	11.06	2352	2.0872	0.6152
0.1881	12.06	2548	2.0095	0.6094
0.0932	13.06	2744	1.9398	0.6232
0.0303	14.06	2940	1.9994	0.6134
0.0513	15.06	3132	1.9757	0.6149

Property	Details
Model Type	A fine - tuned version of google/vivit-b-16x2-kinetics400 on UCF - Crime dataset
Training Data	UCF - Crime dataset
Metrics	Accuracy
Model Name	vivit-b-16x2-kinetics400-UCF-Crime

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base