inference.yml

MODEL:
  PRETRAIN_CHOICE: 'self'
  PRETRAIN_PATH: '/path/to/.cache/torch/checkpoints/resnet50-19c8e357.pth' # from https://download.pytorch.org/models/resnet50-19c8e357.pth
  CENTER_LOSS: 'on'
  CENTER_FEAT_DIM: 2048
  NAME: 'resnet50_nl'
  WEIGHT_REGULARIZED_TRIPLET: 'on'
  GENERALIZED_MEAN_POOL: 'on'
  REDUCE_DIM: 'off'
  REDUCED_DIM: 2048

INPUT:
  IMG_SIZE: [192, 192]
  PROB: 0.5 # random horizontal flip
  RE_PROB: 0.0 # random erasing
  PADDING: 10

DATASETS:
  NAMES: ('combined_tools_redwood_reid_v1')

DATALOADER:
  PK_SAMPLER: 'on'
  NUM_INSTANCE: 4
  NUM_WORKERS: 8
  CLASS_SAMPLER: 'on'
  NUM_CLASS : 16

SOLVER:
  OPTIMIZER_NAME: 'Adam'
  MAX_EPOCHS: 12
  BASE_LR: 0.000035

  CENTER_LR: 0.5
  CENTER_LOSS_WEIGHT: 0.0005

  WEIGHT_DECAY: 0.0005
  IMS_PER_BATCH: 64

  STEPS: [7, 10]
  GAMMA: 0.1

  WARMUP_FACTOR: 0.1
  WARMUP_ITERS: 1
  WARMUP_METHOD: 'linear'

  CHECKPOINT_PERIOD: 1  # in epochs
  LOG_PERIOD: 20        # in iterations
  EVAL_PERIOD: 1        # in epochs

TEST:
  IMS_PER_BATCH: 128
  RE_RANKING: 'off'
  WEIGHT: "./log/Sampling/co3d_reid_v10/v29/resnet50_nl_checkpoint_32532.pt"
  FEAT_NORM: 'on'
  EVALUATE_ONLY: 'off'
  PARTIAL_REID: 'off'
  SAVE_MEMORY: 'on'
  SAVE_MEMORY_BLOCK: 4096

INFERENCE:
 DO_INFERENCE: 'on'
 SAMPLES: 51
 NUM_RETRIEVE: 15
 OUTPUT: "inference_res.json"

WANDB:
 LOG_WANDB: 'off'
 PROJECT: "Sampling"
 NOTES: "Examination of influence of sampling techniques."
 TAGS: ["test", "co3d_reid_v1"]

OUTPUT_DIR: "."
OUTPUT_VERSIONING: 'off'