16_32_prompting_ivlp_s2.yaml

DATA:
    ROOT: '/PATH/TO/videos/folder'
    TRAIN_FILE: 'datasets_splits/base2novel_splits/ssv2_splits/fs_base_s2_train.txt'
    VAL_FILE: 'datasets_splits/base2novel_splits/ssv2_splits/base_val.txt'
    DATASET: ssv2
    NUM_FRAMES: 32
    NUM_CLASSES: 87
    LABEL_LIST: 'labels/ssv2_174_base_labels.csv'
MODEL:
    ARCH: ViT-B/16
TRAIN:
    BATCH_SIZE: 4 # BS 64
    ACCUMULATION_STEPS: 2
    EPOCHS: 11
    LR: 4.e-02
TRAINER:
  ViFi_CLIP:
    PROMPT_MODEL: True # If true, will only train the prompts while keeping rest of the model frozen
    N_CTX_VISION: 16
    N_CTX_TEXT: 16
    CTX_INIT: "a photo of a"
    PROMPT_DEPTH_VISION: 9
    PROMPT_DEPTH_TEXT: 9
SAVE_FREQ: 10