object-detection-react icon indicating copy to clipboard operation
object-detection-react copied to clipboard

Incorrect run test after tranning within google colab

Open trumpiter-max opened this issue 4 years ago • 7 comments

Hi I trained model which is about hand poses on google colab, but it seems not work correctlly. What should I do now?

trumpiter-max avatar Jan 14 '21 16:01 trumpiter-max

What is the error?

bourdakos1 avatar Jan 14 '21 17:01 bourdakos1

This is the input image that I made on https://cloud.annotations.ai/. image Moreover, I tried other poses but all predict result is letter K. image

trumpiter-max avatar Jan 15 '21 07:01 trumpiter-max

Maybe you need more training data? You have a lot of classes that probably all look very similar, so you’ll need more training data for it to be able to distinguish

bourdakos1 avatar Jan 15 '21 15:01 bourdakos1

Sure, I wonder how many images for each class that I need to add?

trumpiter-max avatar Jan 15 '21 15:01 trumpiter-max

I can’t say for sure, but try to get to 250 each and see if the accuracy improves

bourdakos1 avatar Jan 15 '21 15:01 bourdakos1

Ok, thanks, I will try. In fact, I used same model for yolov5 and it runs correctly almost cases.

trumpiter-max avatar Jan 15 '21 16:01 trumpiter-max

Good to know, you could try playing around with the hyperparameters or other models to see if you can improve accuracy. The colab training process is new so there could be an issue on my end, I will look into it

bourdakos1 avatar Jan 15 '21 17:01 bourdakos1