PixelLib VS sahi

This is the code, found from the reference here.

Hi All, I am trying to detect defects in the images using YOLOv8where some of the classes (defectType1, defectType2) have very small bounding boxes and some of them have large bounding boxes associated with the, (defectType3, defectType4). Also, real-time operation is desired (at least 5Hz on Jetson Xavier) What I have done till now: I am primarily trying to use the SAHI technique (Slicing Aided Hyper Inference)

I am using the default YOLOv5m6 model here with sahi/yolov5 library for my object detection project. I want to change just some of labels - for example when YOLO detects a human, I want it to label the human as "threat", not "person". Is there any way I can do it just changing some code, or I should train the model from scratch by just changing labels?

I need to execute this model https://github.com/obss/sahi upon an HTTP request. I will need between 32GB and 128GB of RAM (depending on the request). Also, I will only receive this request once or twice a week (they are not predefined dates). Each process may take a few hours.

https://github.com/obss/sahi should do the job

I think you have multiple options: why run inference on this large resolution? Why not run on 1024x512 or smaller. Use a smaller model which uses less memory, eg enet, erfnet, bisenet etc. Otherwise, patchbased inference is the way to go, there is a nice library, but also easy to implement yourself: https://github.com/obss/sahi

i have the EXACT thing ! the libs github!

Good idea. I haven’t looked too closely yet at the “hard” datasets.

We originally considered “fixing” the labels on these datasets by hand, but ultimately decided that label error is one of the challenges “real world” datasets have that models should work to become more robust against. There is some selection bias in that we did make sure that the datasets we chose passed the eye test (in other words, it looked like the user spent a considerable amount of time annotating & a sample of the images looked like they labeled some object of interest).

For aerial images in particular my guess would be that these models suffer from the “small object problem”[1] where the subjects are tiny compared to the size of the image. Trying a sliding window based approach like SAHI[2] on them would probably produce much better results (at the expense of much lower inference speed).

[1] https://blog.roboflow.com/detect-small-objects/

[2] https://github.com/obss/sahi

I am not very experienced, but do I understand that the problem is the size of the image? If so, have you heard of sahi

Try slicing and yolov4. https://github.com/obss/sahi