Skip to content

a10ra/K3-Computer-Vision-Project

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

9 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

K3-Computer-Vision-Project

Uses the dataset from Roboflow with Yolov11.

How to Run

  1. Use docker to containerize the image (Image is published on DockerHub as a10ra/obj-det-image or on this link https://hub.docker.com/repository/docker/a10ra/obj-det-image/general)
  2. Run the container on the port like: docker run -d --name obj-det-container -p 127.0.0.1:5000:5000 obj-det-image

Problems with the dataset

One primary issue of the dataset is the imbalance. Mask and vest classes are under represented in the dataset.

I attempted to mitigate the impact with data augmentation through albumentations and simpler techniques like oversampling the smaller class.

However, what worked best was using the Weighted Dataloader provided by Yasin's solution

Class Images Instances Box (P) R mAP50 mAP50-95
all 15 49 0.943 0.834 0.9 0.657
helmet 14 21 1 0.988 0.995 0.741
mask 7 14 1 0.657 0.897 0.61
vest 9 14 0.828 0.857 0.808 0.62

Problems I Faced

I faced issues during the model training, where I struggled to get the mean Average Precision to be over 0.5. Trying out different data augmentation, class balancing methods, was also a learning experience for me on this dataset.

Another problem faced was just general build issues - things would work locally but the container wouldn't start or the requirements.txt was missing some dependencies. Overall, I spent a lot more time looking through the documentation and guides from FastAPI, Docker and YOLO.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published