Hubert2102
commented
2 years ago when i try to train the model with all polygons, like 'segmentation': [[342.0, 787.5, 333.0, 784.5, 322.5, 770.0, 325.5, 741.0, 343.5, 680.0, 354.5, 611.0, 349.5, 575.0, 333.5, 567.0, 332.0, 553.5, 305.0, 551.5, 280.0, 541.5, 279.5, 530.0, 304.5, 415.0, 328.5, 335.0, 354.5, 268.0, 347.5, 261.0, 347.5, 254.0, 356.5, 250.0, 353.5, 211.0, 350.5, 193.0, 345.5, 188.0, 346.5, 185.0, 338.5, 170.0, 332.0, 163.5, 292.0, 146.5, 272.0, 134.5, 258.5, 122.0, 255.5, 110.0, 263.0, 100.5, 276.0, 92.5, 349.0, 67.5, 354.5, 61.0, 354.5, 41.0, 365.0, 24.5, 408.0, 24.5, 428.0, 34.5, 440.5, 48.0, 446.5, 60.0, 450.5, 74.0, 449.5, 90.0, 460.5, 100.0, 462.5, 121.0, 465.5, 124.0, 462.5, 142.0, 489.0, 151.5, 500.5, 163.0, 501.5, 186.0, 498.5, 220.0, 495.5, 220.0, 493.5, 289.0, 482.5, 367.0, 480.5, 421.0, 475.5, 439.0, 467.5, 446.0, 459.5, 462.0, 456.5, 568.0, 453.0, 569.5, 444.0, 565.5, 440.0, 574.5, 398.5, 579.0, 385.5, 621.0, 386.5, 662.0, 399.5, 706.0, 391.5, 735.0, 392.5, 746.0, 381.0, 756.5, 368.0, 757.5, 361.0, 752.5, 356.5, 778.0, 342.0, 787.5], [369.5, 142.0, 378.5, 138.0, 369.5, 128.0, 366.0, 76.5, 324.0, 96.5, 295.5, 115.0, 333.0, 134.5, 351.0, 135.5, 369.5, 142.0], [463.5, 385.0, 468.5, 369.0, 468.5, 231.0, 466.0, 229.5, 446.5, 277.0, 455.5, 311.0, 463.5, 385.0], [362.5, 730.0, 370.5, 692.0, 367.0, 679.5, 360.5, 712.0, 362.5, 730.0]]-four polygons which represent the person the inference will get a result which have no Blank hole
i have the images and masks, and can get the "BitMasks", but transfiner need another groundtruth,that`s poly_masks https://github.com/SysCV/transfiner/blob/5b61fb53d8df5484f44c8b7d8415f398fd283ddc/detectron2/data/detection_utils.py#L437 i dont know how to express the instances with mask_format=='polygon'. my mask like the pic below: