Thanks for this video! When you train the custom model, does it get rid of the default object detection? What's the best way to ensure we keep the pre-trained model and just append the custom objects to it?
Hi, i could not see the files best.pt in weights, can you help? Inside weights were 2 zip folders: best and last. I extracted them but could not find the file best.pt anywhere. Im on linux.
Hello! I have some sort of a bug when I train my model. I have a 4VRAM gpu and I receive an error that I do not have enough space, even if i reduce the size of the batch and using the nano yolo version. Then i moved to cpu, having 24gb ram, and after 2/3 epochs the training stops randomly without any warning. Any idea why this happens?
@@TheCodingBug Regarding to cuda error, is there a gradient accumulation argument to use. I don't think training with a single sample per optim step is fine. The problem with sudden stopping is not yielding any kind of error or warning. The last thing i see in the console is the epoch progress and then the path of the program on the next line:). Some said it might be from lack of memory but 24gb should be more than enough for such a small yolo. Maybe i'll go on colab even if i hate it.
This Video was amazing. I like how you explain it, like I was 4years old. Thank you very much you deserve like and subscribe
I got it working right away, thank you!
I am glad it worked out fine.
@@TheCodingBug I have cuda 11.8 version but it saying false
Is it mandatory to install 12.1 version?
Thanks for the tutorials !
very useful for the initiatives
Thanks for this video! When you train the custom model, does it get rid of the default object detection? What's the best way to ensure we keep the pre-trained model and just append the custom objects to it?
Yes it will not detect original classes. You'll need to meege your data annotations with oroginal coco data annotations to achieve this.
Bro can yo do a video how to generate graph from the trained model
Hi, i could not see the files best.pt in weights, can you help? Inside weights were 2 zip folders: best and last. I extracted them but could not find the file best.pt anywhere. Im on linux.
I trained a small dataset (53 samples) of one egg in different positions and places, but never detected any of them during prediction process
UPDATE: it's worked after adding the parameter amp=False to train function and trained the model again
Hello! I have some sort of a bug when I train my model. I have a 4VRAM gpu and I receive an error that I do not have enough space, even if i reduce the size of the batch and using the nano yolo version. Then i moved to cpu, having 24gb ram, and after 2/3 epochs the training stops randomly without any warning. Any idea why this happens?
I am not sure about sudden training stop. Can you share the error?
About "out of cuda memory" error, reduce the btch size to 1.
@@TheCodingBug Regarding to cuda error, is there a gradient accumulation argument to use. I don't think training with a single sample per optim step is fine.
The problem with sudden stopping is not yielding any kind of error or warning. The last thing i see in the console is the epoch progress and then the path of the program on the next line:). Some said it might be from lack of memory but 24gb should be more than enough for such a small yolo. Maybe i'll go on colab even if i hate it.
how I use this with tensorrt ? quand why some peoples use pytorch or tensorflow for this
You can export the model to tensorRT via model.export.
Bro pls do video for amd🙏🙏🙏 i mean rx6700xt or others/// I know that is possible via ROCm but i dont know ho to do it:///
I will try.
@@TheCodingBug thank you!
Can i have ur images and video for training and test?
The trained model, code, images, video is available for our patreon supporters.
How I contact you
You can contact me via LinkedIn. The link is pinned on my channel home page.