debug.yaml

MODEL:
  WEIGHT: ""
  ARCHITECTURE: "CLIP"
  CLIP:
    NUM_CLIPS: 64
    FEATPOOL:
      INPUT_SIZE: 4096
      HIDDEN_SIZE: 512
      KERNEL_SIZE: 4
    VIDEOEMBED:
      PRETRAIN: True
      TRAINABLE: True
    TEXTEMBED:
      PRETRAIN: True
      TRAINABLE: True
    FEAT2D:
      POOLING_COUNTS: [15,8,8]
    INTEGRATOR:
      QUERY_HIDDEN_SIZE: 512
      LSTM:
        NUM_LAYERS: 3
        BIDIRECTIONAL: False
    PREDICTOR:
      HIDDEN_SIZE: 512
      KERNEL_SIZE: 9
      NUM_STACK_LAYERS: 4
    LOSS:
      MIN_IOU: 0.5
      MAX_IOU: 1.0
DATASETS:
  TRAIN: ("activitynet_train", "activitynet_val")
  TEST: ("activitynet_test",)
INPUT:
  NUM_SEGMENTS: 1
  FRAMES_PER_SEGMENT: 16
DATALOADER:
  NUM_WORKERS: 0
SOLVER:
  LR: 0.0005
  BATCH_SIZE: 4
  MILESTONES: (3, 4)
  MAX_EPOCH: 5
TEST:
  NMS_THRESH: 0.5