search

dvlab-research / PanopticFCN

Fully Convolutional Networks for Panoptic Segmentation (CVPR2021 Oral)
Apache License 2.0
391 stars 53 forks source link

Suspicious inference #43

Open EnnioEvo opened 2 years ago

EnnioEvo commented 2 years ago

Thank you for this amazing work.

Inference with pretrained model python3 demo/demo.py --config-file projects/PanopticFCN/configs/PanopticFCN-R50-600-3x-FAST.yaml --input ~/Documents/Ennio/datasets/selection/m3-2b_20210516_191154_diff.jpg gives me this suspicious result:

det1

Where the whole images is detected as a single instance (and is even wrong) Result is the same with different images and different yaml config files

Any clue on what can be?

yanwei-li commented 2 years ago

Hi, sorry for the late reply. It seems strange. Are all the inputs got the same results as this case? Or this is the only case?

sidney001 commented 2 years ago

Same suspicious result. The code I ran in shell is python3 demo/demo.py --config-file projects/PanopticFCN/configs/PanopticFCN-R50-3x.yaml \ --input images/*.jpg --output results/PanopticFCN --opts MODEL.WEIGHTS weights/panoptic_fcn_r50_3x.pth . 16004479832_a748d55f21_k 17790319373_bd19b24cfc_k Transsioner20220616-191730

Or it is a detectron2 version question?

sidney001 commented 1 year ago

@yanwei-li Look forward to your replies