Help in dataset for object detection in videos

[mo_t1]

Hi all I am a student , and I have this Object Detection Project on video, i really need help I am so depressed I almost never sleep for last week , i am trying to solve the problem with my Weapon Dataset Its Dataset for detecting weapons in videos, i did everything i can , annotating with bound box almost 4000 images and still, when i train in colab, the detection is very very bad whether in images or in video its disastor, if you have any ideas i really appreciate it

like this image it either give me very low confince or doesnt even detect it or give me in the bound box another object class,

[Mohamed]

Hi @mo_t1 - you will want to include more images within your project that resemble the images or video you would like to run inference (tests) on with a deployed model.

Gun detection is difficult as guns come in many shapes, sizes, and even colors. Don’t feel depressed. You’re on the right track!

More resources to help with improving your dataset and run inference with video:

• https://help.roboflow.com/dataset-health-check-improving-your-dataset-

• https://help.roboflow.com/health-check-filter-by-class

• Video inference (custom YOLOv5 model): https://help.roboflow.com/run-yolov5-on-your-webcam
  

  

• Video inference (Roboflow): GitHub - roboflow/video-inference: Example showing how to do inference on a video file with Roboflow Infer

Here’s more context on the process to implement active learning to improve your dataset:

2 bound boxes for same object

Hi @mo_t1, it looks like your model has two classes it has trouble differentiating between. I recommend implementing Active Learning and taking more examples of the PKS and RPG classes, as well as this example image, and from there: upload them to Roboflow Annotate the images Generate a new version of your dataset Retrain your model Retest your model on a similar image, and more images, and repeat the process to root out errors and raise the confidence level on all detections