could you provide V19-Lite trained models for Panoptic segmentation

[griffintin]

Hi, @youngwanLEE

Thank you so much for sharing this amazing work. I am about to evaluate centermask for Panoptic segmentation, and mainly focused on low-power consumption models.

Is it possible to share models trained on V19-Lite-Slim architecture for Panoptic-Segmentation? V39 is really good, but i am wondering for a more compact network.

Thank you for your time and hopping to hearing from you.

[griffintin]

@youngwanLEE
hello, I trained V19-Lite-Slim for Panoptic-Segmentation, However, accuracy for mask AP is 24.2, bbox AP is 26.7 and PQ is 32.02. Since there is no provided data in your Git or in the paper for this architecture, I refer to mask AP in instance segmentation using the same architecture, which is 29.80.

you can see, mask AP in my own panoptic training is not as good as that in your provided instance segmentation. I have no idea why, can you share any ideas.

By the way, I disabled mask_scoring, which takes too many parameters.

Here is the config file:

CUDNN_BENCHMARK: false DATALOADER: ASPECT_RATIO_GROUPING: true FILTER_EMPTY_ANNOTATIONS: true NUM_WORKERS: 4 REPEAT_THRESHOLD: 0.0 SAMPLER_TRAIN: TrainingSampler DATASETS: PRECOMPUTED_PROPOSAL_TOPK_TEST: 1000 PRECOMPUTED_PROPOSAL_TOPK_TRAIN: 2000 PROPOSAL_FILES_TEST: [] PROPOSAL_FILES_TRAIN: [] TEST:

• coco_2017_val_panoptic_separated TRAIN:
• coco_2017_train_panoptic_separated GLOBAL: HACK: 1.0 INPUT: CROP: ENABLED: false SIZE:
  • 0.9
  • 0.9 TYPE: relative_range FORMAT: BGR MASK_FORMAT: polygon MAX_SIZE_TEST: 1000 MAX_SIZE_TRAIN: 1000 MIN_SIZE_TEST: 600 MIN_SIZE_TRAIN:
• 580
• 600 MIN_SIZE_TRAIN_SAMPLING: choice MODEL: ANCHOR_GENERATOR: ANGLES:
  • • -90
    • 0
    • 90 ASPECT_RATIOS:
  • • 0.5
    • 1.0
    • 2.0 NAME: DefaultAnchorGenerator OFFSET: 0.0 SIZES:
  • • 32
    • 64
    • 128
    • 256
    • 512 BACKBONE: FREEZE_AT: 0 NAME: build_fcos_vovnet_fpn_backbone DEVICE: cuda FCOS: CENTER_SAMPLE: true FPN_STRIDES:
  • 8
  • 16
  • 32
  • 64
  • 128 INFERENCE_TH_TEST: 0.05 INFERENCE_TH_TRAIN: 0.05 IN_FEATURES:
  • p3
  • p4
  • p5
  • p6
  • p7 LOC_LOSS_TYPE: giou LOSS_ALPHA: 0.25 LOSS_GAMMA: 2.0 NMS_TH: 0.6 NORM: GN NUM_BOX_CONVS: 2 NUM_CLASSES: 80 NUM_CLS_CONVS: 2 NUM_SHARE_CONVS: 0 POST_NMS_TOPK_TEST: 50 POST_NMS_TOPK_TRAIN: 100 POS_RADIUS: 1.5 PRE_NMS_TOPK_TEST: 1000 PRE_NMS_TOPK_TRAIN: 1000 PRIOR_PROB: 0.01 SIZES_OF_INTEREST:
  • 64
  • 128
  • 256
  • 512 THRESH_WITH_CTR: false TOP_LEVELS: 2 USE_DEFORMABLE: false USE_RELU: true USE_SCALE: true FPN: FUSE_TYPE: sum IN_FEATURES:
  • stage2
  • stage3
  • stage4
  • stage5 NORM: '' OUT_CHANNELS: 128 KEYPOINT_ON: false LOAD_PROPOSALS: false MASKIOU_LOSS_WEIGHT: 1.0 MASKIOU_ON: false MASK_ON: true META_ARCHITECTURE: PanopticFPN MOBILENET: false PANOPTIC_FPN: COMBINE: ENABLED: true INSTANCES_CONFIDENCE_THRESH: 0.5 OVERLAP_THRESH: 0.5 STUFF_AREA_LIMIT: 4096 INSTANCE_LOSS_WEIGHT: 1.0 PIXEL_MEAN:
• 103.53
• 116.28
• 123.675 PIXEL_STD:
• 1.0
• 1.0
• 1.0 PROPOSAL_GENERATOR: MIN_SIZE: 0 NAME: FCOS RESNETS: DEFORM_MODULATED: false DEFORM_NUM_GROUPS: 1 DEFORM_ON_PER_STAGE:
  • false
  • false
  • false
  • false DEPTH: 50 NORM: FrozenBN NUM_GROUPS: 1 OUT_FEATURES:
  • res4 RES2_OUT_CHANNELS: 256 RES5_DILATION: 1 STEM_OUT_CHANNELS: 64 STRIDE_IN_1X1: true WIDTH_PER_GROUP: 64 RETINANET: BBOX_REG_WEIGHTS:
  • 1.0
  • 1.0
  • 1.0
  • 1.0 FOCAL_LOSS_ALPHA: 0.25 FOCAL_LOSS_GAMMA: 2.0 IN_FEATURES:
  • p3
  • p4
  • p5
  • p6
  • p7 IOU_LABELS:
  • 0
  • -1
  • 1 IOU_THRESHOLDS:
  • 0.4
  • 0.5 NMS_THRESH_TEST: 0.5 NUM_CLASSES: 80 NUM_CONVS: 4 PRIOR_PROB: 0.01 SCORE_THRESH_TEST: 0.05 SMOOTH_L1_LOSS_BETA: 0.1 TOPK_CANDIDATES_TEST: 1000 ROI_BOX_CASCADE_HEAD: BBOX_REG_WEIGHTS:
  • • 10.0
    • 10.0
    • 5.0
    • 5.0
  • • 20.0
    • 20.0
    • 10.0
    • 10.0
  • • 30.0
    • 30.0
    • 15.0
    • 15.0 IOUS:
  • 0.5
  • 0.6
  • 0.7 ROI_BOX_HEAD: BBOX_REG_WEIGHTS:
  • 10.0
  • 10.0
  • 5.0
  • 5.0 CLS_AGNOSTIC_BBOX_REG: false CONV_DIM: 256 FC_DIM: 1024 NAME: '' NORM: '' NUM_CONV: 0 NUM_FC: 0 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_TYPE: ROIAlignV2 SMOOTH_L1_BETA: 0.0 TRAIN_ON_PRED_BOXES: false ROI_HEADS: BATCH_SIZE_PER_IMAGE: 512 IN_FEATURES:
  • p3
  • p4
  • p5 IOU_LABELS:
  • 0
  • 1 IOU_THRESHOLDS:
  • 0.5 NAME: CenterROIHeads NMS_THRESH_TEST: 0.5 NUM_CLASSES: 80 POSITIVE_FRACTION: 0.25 PROPOSAL_APPEND_GT: true SCORE_THRESH_TEST: 0.05 ROI_KEYPOINT_HEAD: ASSIGN_CRITERION: ratio CONV_DIMS:
  • 512
  • 512
  • 512
  • 512
  • 512
  • 512
  • 512
  • 512 IN_FEATURES:
  • p2
  • p3
  • p4
  • p5 LOSS_WEIGHT: 1.0 MIN_KEYPOINTS_PER_IMAGE: 1 NAME: KRCNNConvDeconvUpsampleHead NORMALIZE_LOSS_BY_VISIBLE_KEYPOINTS: true NUM_KEYPOINTS: 17 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_TYPE: ROIAlignV2 ROI_MASKIOU_HEAD: CONV_DIM: 128 NAME: MaskIoUHead NUM_CONV: 2 ROI_MASK_HEAD: ASSIGN_CRITERION: ratio CLS_AGNOSTIC_MASK: false CONV_DIM: 128 NAME: SpatialAttentionMaskHead NORM: '' NUM_CONV: 2 POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 0 POOLER_TYPE: ROIAlignV2 RPN: BATCH_SIZE_PER_IMAGE: 256 BBOX_REG_WEIGHTS:
  • 1.0
  • 1.0
  • 1.0
  • 1.0 BOUNDARY_THRESH: -1 HEAD_NAME: StandardRPNHead IN_FEATURES:
  • res4 IOU_LABELS:
  • 0
  • -1
  • 1 IOU_THRESHOLDS:
  • 0.3
  • 0.7 LOSS_WEIGHT: 1.0 NMS_THRESH: 0.7 POSITIVE_FRACTION: 0.5 POST_NMS_TOPK_TEST: 1000 POST_NMS_TOPK_TRAIN: 2000 PRE_NMS_TOPK_TEST: 6000 PRE_NMS_TOPK_TRAIN: 12000 SMOOTH_L1_BETA: 0.0 SEM_SEG_HEAD: COMMON_STRIDE: 4 CONVS_DIM: 128 IGNORE_VALUE: 255 IN_FEATURES:
  • p2
  • p3
  • p4
  • p5 LOSS_WEIGHT: 0.5 NAME: SemSegFPNHead NORM: GN NUM_CLASSES: 54 VOVNET: BACKBONE_OUT_CHANNELS: 256 CONV_BODY: V-19-slim-eSE DEFORMABLE_GROUPS: 1 NORM: FrozenBN OUT_CHANNELS: 256 OUT_FEATURES:
  • stage2
  • stage3
  • stage4
  • stage5 STAGE_WITH_DCN:
  • false
  • false
  • false
  • false WITH_MODULATED_DCN: false WEIGHTS: output/Panoptic-CenterMask-V-19-slim-eSE-FPN-ms-3x/model_final.pth OUTPUT_DIR: output/Panoptic-CenterMask-V-19-slim-eSE-FPN-ms-3x SEED: -1 SOLVER: BASE_LR: 0.005 BIAS_LR_FACTOR: 1.0 CHECKPOINT_PERIOD: 10000 CLIP_GRADIENTS: CLIP_TYPE: value CLIP_VALUE: 1.0 ENABLED: false NORM_TYPE: 2.0 GAMMA: 0.1 IMS_PER_BATCH: 8 LR_SCHEDULER_NAME: WarmupMultiStepLR MAX_ITER: 720000 MOMENTUM: 0.9 NESTEROV: false STEPS:
• 500000
• 600000 WARMUP_FACTOR: 0.001 WARMUP_ITERS: 1000 WARMUP_METHOD: linear WEIGHT_DECAY: 0.0001 WEIGHT_DECAY_BIAS: 0.0001 WEIGHT_DECAY_NORM: 0.0 TEST: AUG: ENABLED: false FLIP: true MAX_SIZE: 4000 MIN_SIZES:
  • 400
  • 500
  • 600
  • 700
  • 800
  • 900
  • 1000
  • 1100
  • 1200 DETECTIONS_PER_IMAGE: 100 EVAL_PERIOD: 0 EXPECTED_RESULTS: [] KEYPOINT_OKS_SIGMAS: [] PRECISE_BN: ENABLED: false NUM_ITER: 200 VERSION: 2 VIS_PERIOD: 0

[griffintin]

@youngwanLEE

I think the results in the table of Panoptic-CenterMask are wrong. The data for box AP and mask AP are opposite.

[abhigoku10]

@griffintin can you please share ur model on the onedrive or google drive ? it would be helpfull for panoptic segmentation