vit-base-patch16-224-in21k_Human_Activity_Recognition Open-source Model - Empowering Image Classification for Human Activity Recognition

Vit Base Patch16 224 In21k Human Activity Recognition

Developed by DunnBC22

A fine-tuned human activity recognition model based on Google Vision Transformer (ViT) for image classification tasks

EnglishOpen Source License:Apache-2.0 #Human Activity Recognition #ViT Fine-tuned Model #Multi-class Image Classification

Downloads 92

Release Time : 1/24/2023

Model Overview

This model is fine-tuned from Google's ViT model, specifically designed to recognize various human activities in images. It achieved an accuracy of 83.81% on the evaluation set.

Model Features

High Accuracy

Achieved 83.81% accuracy on human activity recognition tasks

Based on ViT Architecture

Fine-tuned using Google Vision Transformer base model

Multi-metric Evaluation

Provides various evaluation metrics including accuracy, F1 score, recall, and precision

Model Capabilities

Image Classification

Human Activity Recognition

Multi-class Classification

Use Cases

Surveillance and Security

Surveillance Video Analysis

Identify human activities in surveillance videos

Can accurately recognize various human activities

Health and Fitness

Exercise Movement Recognition

Identify movements during fitness or sports activities

🚀 vit-base-patch16-224-in21k_Human_Activity_Recognition

This model is a fine - tuned version of google/vit-base-patch16-224-in21k. It's designed for human activity recognition through image classification, offering a practical solution for related tasks.

📚 Documentation

Model Information

Property	Details
Model Type	Fine - tuned version of google/vit-base-patch16-224-in21k
Training Data	imagefolder
Metrics	Accuracy, F1, Recall, Precision

Model Performance

It achieves the following results on the evaluation set:

Loss: 0.7403
Accuracy: 0.8381
F1
- Weighted: 0.8388
- Micro: 0.8381
- Macro: 0.8394
Recall
- Weighted: 0.8381
- Micro: 0.8381
- Macro: 0.8390
Precision
- Weighted: 0.8421
- Micro: 0.8381
- Macro: 0.8424

Model description

This is a multiclass image classification model of humans doing different activities.

For more information on how it was created, check out the following link: https://github.com/DunnBC22/Vision_Audio_and_Multimodal_Projects/blob/main/Computer%20Vision/Image%20Classification/Multiclass%20Classification/Human%20Activity%20Recognition/ViT-Human%20Action_Recogniton.ipynb

Intended uses & limitations

This model is intended to demonstrate my ability to solve a complex problem using technology. You are welcome to test and experiment with this model, but it is at your own risk/peril.

Training and evaluation data

Dataset Source: https://www.kaggle.com/datasets/meetnagadia/human-action-recognition-har-dataset

Sample Images From Dataset:

Sample Images

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.0002
train_batch_size: 16
eval_batch_size: 8
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e - 08
lr_scheduler_type: linear
num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Weighted F1	Micro F1	Macro F1	Weighted Recall	Micro Recall	Macro Recall	Weighted Precision	Micro Precision	Macro Precision
1.0814	1.0	630	0.7368	0.7794	0.7795	0.7794	0.7798	0.7794	0.7794	0.7797	0.7896	0.7794	0.7896
0.5149	2.0	1260	0.6439	0.8060	0.8049	0.8060	0.8036	0.8060	0.8060	0.8051	0.8136	0.8060	0.8130
0.3023	3.0	1890	0.7026	0.8254	0.8272	0.8254	0.8278	0.8254	0.8254	0.8256	0.8335	0.8254	0.8345
0.0507	4.0	2520	0.7414	0.8317	0.8342	0.8317	0.8348	0.8317	0.8317	0.8321	0.8427	0.8317	0.8438
0.0128	5.0	3150	0.7403	0.8381	0.8388	0.8381	0.8394	0.8381	0.8381	0.8390	0.8421	0.8381	0.8424

Framework versions

Transformers 4.25.1
Pytorch 1.12.1
Datasets 2.8.0
Tokenizers 0.12.1

📄 License

This model is licensed under the Apache - 2.0 license.

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご