Training with different sizes sliced images?

[AI-ML-Enthusiast]

@avanetten I would like to train different sizes of sliced images. Not only 416, or 544. It maybe mixed with 416 and 544, or other. How can I do it? Would you suggest me please?

[Neykah]

If you plan to use yolo v3, this is already implemented, as it often proves itself a meaningful data augmentation step. In the [yolo] parts of the config file, there should be a line which states: random=1 If so, your input patches are resized to a random size between 320x320 and 608x608 every 10 iterations. If you want to do something more specific however, I guess you will have to modify the C code.

[AI-ML-Enthusiast]

@Neykah @avanetten Thank you very much for your comment.

If you plan to use yolo v3, this is already implemented, as it often proves itself a meaningful data augmentation step. In the [yolo] parts of the config file, there should be a line which states: random=1

Since my training images are not fixed size, what about the parameters "train_input_width" and "train_input_height" in yolt3? Also what about the config file parameters width and height?

Thanks

[AI-ML-Enthusiast]

@Neykah @avanetten It is training with different size using yolt3. I changed to maximum size of the train_input_width, train_input_height in simrdwn.py, and width ,height parameters in config file. I don't know it is OK or not, but it is training with those images. Thanks