Queries on training custom dataset for panoptic task

[sahityabonumaddi]

I know these are very basic queries, thanks in advance !

1. After writing a file similar to https://github.com/SHI-Labs/OneFormer/blob/main/oneformer/data/datasets/register_ade20k_panoptic.py to register custom data, I should run this specific python file separately to register. is that right?

2. I'm using oneformerunifieddatasetmapper https://github.com/SHI-Labs/OneFormer/blob/main/oneformer/data/dataset_mappers/oneformer_unified_dataset_mapper.py

3. I'm using config file similar to https://github.com/SHI-Labs/OneFormer/blob/5e04c9aaffd9bc73020d2238757f62346fe778c0/configs/ade20k/Base-ADE20K-UnifiedSegmentation.yaml

4.I have doubt in evaluator, can I use any provided evaluators for my dataset? if not kindly guide me through

[praeclarumjj3]

Hi @sahityabonumaddi, thanks for your interest in our work.

After writing a file similar to https://github.com/SHI-Labs/OneFormer/blob/main/oneformer/data/datasets/register_ade20k_panoptic.py to register custom data, I should run this specific python file separately to register. is that right?

You don't need to run that file. Just import it inside oneformer/data/datasets/init.py.

I have doubt in evaluator, can I use any provided evaluators for my dataset? if not kindly guide me through

I am sorry, but I cannot answer that question with no knowledge about the structure of your dataset. if it's in COCO format, then you can use COCOPanopticEvaluator. Otherwise, you might need to write your own evaluator.

[sahityabonumaddi]

Except run length coding(RLE) in segmenting a object, it is in detectron2 standard format.

[sahityabonumaddi]

once I register my dataset in detectron2 standard format, which is done please guide me through how to write a evaluator for custom dataset @praeclarumjj3

[praeclarumjj3]

Hi @sahityabonumaddi, Could you share the JSON format for your panoptic annotations JSON file? it is unclear to me what's the format of your dataset without which I can't help you.

Except run length coding(RLE) in segmenting a object, it is in detectron2 standard format.

There's no standard detectron2 format for a dataset. Did you mean COCO format?

[sahityabonumaddi]

Thanks for your reply @praeclarumjj3

By standard detectron2 standard format I mean the format of list[dict] as specified by https://detectron2.readthedocs.io/en/latest/tutorials/datasets.html

I modified the def load_ade20k_panoptic_json(json_file, image_dir, gt_dir, semseg_dir, meta): function in register_ade20k_panoptic.py to return the specified keys.

ret.append( { "file_name": image_file, "image_id": image_id, "pan_seg_file_name": label_file, "sem_seg_file_name": sem_label_file, "segments_info": segments_info, } )

this is my JSON file for panoptic annotations.

{"id": "00001", "width": 1024, "height": 1024, "file_name": "images/airport/100/img_1_0_1552039977070010400.png" ,

"detection": [{"id": 3853443, "category_id": 13, "category_name": "smallvehicle", "segmentation": [459, 985, 458, 986, 457, 986, 456, 986, 455, 987, 454, 987, 453, 987, 453, 988, 453, 989, 453, 990, 453, 991, 453, 992, 453, 993, 453, 994, 453, 995, 453, 996, 454, 997, 455, 997, 456, 997, 457, 997, 458, 997, 459, 998, 460, 998, 461, 998, 462, 998, 463, 998, 464, 998, 465, 998, 466, 998, 467, 998, 468, 998, 469, 998, 470, 998, 471, 997, 472, 997, 473, 997, 474, 997, 475, 997, 476, 997, 477, 996, 478, 996, 479, 995, 480, 994, 480, 993, 480, 992, 480, 991, 480, 990, 480, 989, 480, 988, 479, 987, 479, 986, 478, 986, 477, 986, 476, 985, 475, 985, 474, 985, 474, 986, 473, 987, 472, 986, 471, 986, 470, 986, 469, 986, 468, 986, 467, 986, 466, 986, 465, 986, 464, 986, 463, 986, 462, 986, 461, 985, 460, 985], "hbbox": [453, 985, 28, 14], "obbox": [[453, 998], [453, 985], [480, 985], [480, 998]]},

{"id": 11965944, "category_id": 13, "category_name": "smallvehicle", "segmentation": [861, 819, 860, 820, 860, 821, 860, 822, 859, 823, 859, 824, 859, 825, 859, 826, 860, 827, 861, 828, 862, 829, 863, 829, 864, 830, 865, 830, 866, 831, 867, 831, 868, 831, 869, 832, 869, 833, 870, 832, 871, 832, 872, 833, 873, 833, 874, 834, 875, 834, 876, 834, 877, 835, 878, 835, 879, 835, 880, 836, 881, 836, 882, 836, 883, 836, 884, 837, 885, 836, 885, 835, 885, 834, 886, 833, 886, 832, 886, 831, 887, 830, 886, 829, 885, 828, 884, 827, 883, 826, 882, 826, 881, 825, 880, 824, 879, 824, 878, 825, 877, 824, 876, 824, 875, 823, 874, 823, 873, 823, 872, 822, 871, 822, 870, 821, 869, 821, 868, 821, 867, 820, 866, 819, 865, 819, 864, 819, 863, 819, 862, 819], "hbbox": [859, 819, 29, 19], "obbox": [[884, 838], [856, 828], [860, 817], [888, 826]]},

{"id": 5370071, "category_id": 13, "category_name": "smallvehicle", "segmentation": [833, 768, 832, 769, 831, 769, 830, 769, 829, 770, 828, 770, 828, 771, 828, 772, 828, 773, 828, 774, 829, 775, 829, 776, 830, 777, 830, 778, 831, 779, 831, 780, 831, 781, 832, 782, 832, 783, 832, 784, 833, 785, 833, 786, 833, 787, 833, 788, 834, 789, 834, 790, 835, 791, 836, 792, 836, 793, 837, 794, 837, 795, 838, 796, 839, 796, 840, 796, 841, 795, 842, 795, 843, 795, 844, 794, 845, 794, 846, 793, 845, 792, 845, 791, 845, 790, 845, 789, 844, 788, 844, 787, 844, 786, 843, 785, 843, 784, 843, 783, 842, 782, 842, 781, 841, 780, 841, 779, 841, 778, 840, 777, 840, 776, 840, 775, 839, 774, 839, 773, 838, 772, 838, 771, 837, 770, 836, 769, 835, 768, 834, 768], "hbbox": [828, 768, 19, 29], "obbox": [[836, 797], [826, 770], [837, 766], [846, 793]]},

{"id": 12301172, "category_id": 13, "category_name": "smallvehicle", "segmentation": [830, 829, 829, 830, 828, 830, 827, 830, 826, 831, 825, 832, 825, 833, 825, 834, 825, 835, 826, 836, 826, 837, 827, 838, 827, 839, 828, 840, 828, 841, 829, 842, 829, 843, 829, 844, 830, 845, 830, 846, 830, 847, 831, 848, 830, 849, 831, 850, 831, 851, 832, 852, 833, 853, 833, 854, 834, 854, 835, 855, 836, 855, 837, 855, 838, 855, 839, 854, 840, 854, 841, 854, 842, 853, 842, 852, 842, 851, 842, 850, 842, 849, 842, 848, 842, 847, 842, 846, 841, 845, 841, 844, 840, 843, 840, 842, 840, 841, 839, 840, 839, 839, 839, 838, 838, 837, 838, 836, 838, 835, 837, 834, 837, 833, 836, 832, 836, 831, 835, 830, 835, 829, 834, 829, 833, 829, 832, 829, 831, 829], "hbbox": [825, 829, 18, 27], "obbox": [[832, 856], [823, 831], [835, 827], [844, 852]]}],

"segmentation": {"panoptic_filename": "panoptic/airport/100/img_1_5_1552039977068509700.png", "semantic_filename": "semantic/airport/100/img_1_5_1552039977068509700.png"}}

thanks in advance!