SyGoing
SyGoing
And,when I change the dataloader to v5loader ,the train labels are correct.
> @SyGoing hi, please provide more info, such as ultralytics package version and your training command. Also please show us the visualized batch data if you don't mind. :) the...
Thanks for your reply, when I delete the old cache and train it again, the labels change correct. And, When I close it, train it again without delete the caches,...
hello I used this code ,but it can detect nothing , and you have meat this problem?
Since the ap's computaion method is different from the star-cloud , Maybe there is nothing to compare. If you are intertested to it ,you could compare them using the same...
@yokings Hi, I have added the pretrained model 's ap on the wilderface. And the centerface's ap is computed with the multi-scale,while I got it by single-scale withe voc's ap
Hi , Maybe you could try to convert the mxnet model to onnx form firstly, then convert onnx to openvino. I have checked it and it works well.
I'll try it for train and forward test, if neccessary I would add new layer or some new implementation