devis_ablation3_increased-spatial-inputs.yaml

MODEL:
  WEIGHTS: 'weights/r50_deformable_detr_segmentation_ablation_1lvl_res.pth'
  MASK_ON: True
  NUM_QUERIES: 60
  MASK_HEAD:
    USE_MDC: False
    UPSAMPLING_RESOLUTIONS: ['/32']
    USED_FEATURES: [['/32', 'encoded'], ['/16', 'backbone'], ['/8', 'backbone'], ['/4', 'backbone']]
    DEVIS:
      CONV_HEAD_3D: True

  MATCHER:
    CLASS_COST: 1.0
    USE_SUM_L1_DISTANCE: False

  DEVIS:
    DEFORMABLE_ATTENTION:
      INSTANCE_AWARE_ATTENTION: False

  LOSS:
    MASK_AUX_LOSS: [ ]

DATASETS:
  TYPE: 'vis'
  TRAIN_DATASET: 'yt_vis_train_19'
  VAL_DATASET: 'yt_vis_val_19'

INPUT:
  MIN_SIZE_TEST: 360
  MAX_SIZE_TEST: 640

SOLVER:
  BASE_LR: 0.0001
  LR_BACKBONE: 0.00001
  EPOCHS: 10
  BATCH_SIZE: 1
  STEPS: [3, 7]

TEST:
  START_EVAL_EPOCH: 3
  NUM_OUT: 10
  CLIP_TRACKING:
    USE_BINARY_MASK_IOU: True
    USE_FRAME_AVERAGE_IOU: True
    CLASS_COST: 0
    SCORE_COST: 0