Real-Time Animated Characters Detection and Recognition with YOLOv5

Examine the performance of the YOLOv5 algorithm in detecting and recognizing Tom, Jerry, and Spike.

Performances

Model	Input Size	Batch Size	Dataset Size	Val [email protected]	Test [email protected]
Nano	640x640	16	25%	0.964	0.989
Nano	640x640	16	50%	0.969	0.953
Nano	640x640	16	75%	0.989	0.982
Nano	640x640	16	100%	0.989	0.977
Small	640x640	16	25%	0.989	0.921
Small	640x640	16	50%	0.983	0.935
Small	640x640	16	75%	0.977	0.98
Small	640x640	16	100%	0.98	0.985
Medium	640x640	16	25%	0.946	0.914
Medium	640x640	16	50%	0.984	0.982
Medium	640x640	16	75%	0.971	0.992
Medium	640x640	16	100%	0.983	0.986

Training Dataset

• Images: 1000
  • Has an object: 931
  • No object: 69
• Object Instances:
  • Tom: 562
  • Spike: 538
  • Jerry: 490
• Split Ratio:
  • Train: 70%
  • Validation: 21%
  • Test: 9%

Testing Dataset

• Nano Models
• Small Models
• Medium Models

Documentation

• Presentation

Hardwares

• CPU: AMD Ryzen 9 5900X
• Memory: G.Skill Trident Z 32GB @ 3600 Mhz
• GPU: EVGA GeForce RTX 3060
• Storage: Western Digital 1TB WD Blue

Built With

• Python
• Torch
• YOLOv5

Collaborators

• Sotheanith Sok

Course

• CECS 553 - Machine Vision

License

This project is licensed under the MIT License - see the LICENSE.md file for details

Acknowledgments

Special thank to

• Best-README-Template - the readme template.