search

yingkunwu / R-YOLOv4

This is a PyTorch-based R-YOLOv4 implementation which combines YOLOv4 model and loss function from R3Det for arbitrary oriented object detection.
114 stars 20 forks source link

Values normalization #63

Closed PrefectSol closed 10 months ago

PrefectSol commented 10 months ago

It is not very clear what the input of the model is. I understand that in labels.txt The following is expected:

<class_id> <centerX/imgsz> <centerY/imgsz> <w/imgsz> <h/imgsz> <theta/360 or 2P?>

What is expected on theta? Or maybe the values don't have to be normalized at all

yingkunwu commented 10 months ago

Hi, which labels.txt are you referring to? The expected data format is depending on different dataset. May I ask which dataset is the one you mentioned?

PrefectSol commented 10 months ago

Hi, which labels.txt are you referring to? The expected data format is depending on different dataset. May I ask which dataset is the one you mentioned?

I don't use ready-made data sets, I'm trying to create my own

PrefectSol commented 10 months ago

Hi, which labels.txt are you referring to? The expected data format is depending on different dataset. May I ask which dataset is the one you mentioned?

I don't use ready-made data sets, I'm trying to create my own

That is, I am trying to create my own dataset, but I do not understand in what format the model is expected to be input

yingkunwu commented 10 months ago

For that you can look into this line. This is what I passed to the model from PyTorch Dataset loader. If you want to create your own dataset, you can follow the workflow of the custom dataset that I created.

PrefectSol commented 10 months ago

thx