dvornikita / context_aug Goto Github PK

How to do if I want to do argument on COCO ，though I have known the project developed on VOC-Dataset.

hello,I have got it.

What should I do to prepare my own dataset ?

As the fig.7 shows, there are many unreasonable synthesis images, are these images used for training the Faster-RCNN or BlitzNet? Do these images hurt the detection performance?

Hello！The parameter “image_size：which image size to chose for training” in config.py is mean to scale the image by "image_size" and then crop the box containing the target,I don't understand its meaning.Can you explain the whole process to me about training the Context Model, for example, after inputting the image, crop the target according to the annotation information, and then send them to the classification network? So the "batch_size” in config.py is the number of objects containing the background?

I got the following error at the inference stage:

FileNotFoundError: [Errno 2] No such file or directory: /path/to/context_aug/Data/context_probs/voc12train_neg3-voc12val_small-1neib/2007_000033.npy'

After training, there is no .npy file appearing in my directories, so this is the reason but I could not understand which part I am missing ?

Thanks.