mega.pytorch

mega.pytorch copied to clipboard

mega_core.trainer ERROR: Iteration=4 || Image names used for training ('train/ILSVRC2015_VID_train_0003/ILSVRC2015_train_00839000/000255',) || targets Length=[0] (I have modified the error log infor to print the image file name)

In self.filter_annotation(), class VIDMEGADataset, we already have filtered out the images with len(objs) == 0

So Why this mega_core.trainer ERROR occurs ？ Has the target ( returned by VIDMEGADataset._get_train() ) been modified somewhere else ?

Looking forward to your answer

I also have the problem, have you solved it ?

I also have the problem, have you solved it ?

It’s been a long time, I probably forgot my previous operation.

in fact I have modified the dataloader and i made some mistakes. in my modified dataloader, I reversed the width and height of class BoxList, and as for the gt_target, after target = target.clip_to_image, it will cause many bboxed outside the image and be clipped, thus result in targets Length=[0]

I also have the problem, have you solved it ?

again, It’s been a long time, I probably forgot my previous operation, perhaps, when I use the original dataloader of MEGA ,it also occurs targets Length=[0], but the amount is much less than that using the dataloader modified by my above mistaken operation.

I also have the problem, have you solved it ?

and, if you use the original dataloader of MAGE and also has this ERROR, then I don't konw why and I haven't solved this.

I also have the problem, have you solved it ?

and, if you use the original dataloader of MAGE and also has this ERROR, then I don't konw why and I haven't solved this.

Thank you for your answers!