Lu Qi

We regard all entities as instances in PanopticFCN and DETR without the modification of the code.

For a fair comparison, there are two settings to train ADE20K including the similar training epoch or similar training iteration to the COCO dataset (12 epochs or 90000 iterations with...

Firstly, the reference code is too slow because there was no maskrcnn-benchmark or mmdetection at that time. I recommend extracting some useful modules into the other frameworks. Secondly, I indeed...

Our network is only implemented in the KINS dataloader. However, it is easy to use COCOA because the annotation format of KINS strictly follows the COCO.

Thanks for your attention. However, I am afraid I cannot find this kind of code... It has been about three years... and I do not own the computer that building...

Hi, I am sorry to reply late. Yes, I indeed modify the cocoapi to train and evaluate amodal and inmodal masks simultaneously. The main modification for COCOAPI is changing the...

Hi, if you want to train an amodal mask, changing the key "amodal" to "segm" in the json file is most convenient way.

You could refer to this code. cvbase is only used to load json file. import cvbase as cvb base_path = "/data/qilu/datasets/amodal/" ann_path = os.path.join(base_path, "train.json") save_path = os.path.join(base_path, "newtrain2017.json") aa...

In my method, I also used the occlusion part of the whole image. I think the best way is to use MMDetection or detectron2. You could only put the multi-view...

Oh, I am sorry that I do not release the occlusion ordering now. Please wait for some days, I will update it.