anet_ret_base_resnet50.json

{
  "train_datasets": [
    {
      "name": "anet",
      "txt": "/txt/anet_retrieval/train.jsonl",
      "img": "/img/anet/"
    }
  ],
  "val_datasets": [
    {
      "name": "anet_retrieval",
      "txt": "/txt/anet_retrieval/dev.jsonl",
      "img": "/img/anet/"
    }
  ],
  "max_txt_len": 100,
  "max_img_size": 448,
  "img_pixel_mean": [123.675, 116.28, 103.53],
  "img_pixel_std": [1.0, 1.0, 1.0],
  "img_input_format": "BGR",
  "fps": 1,
  "num_frm": 2,
  "train_n_clips": 4,
  "score_agg_func": "lse",
  "max_n_example_per_group": 1,
  "model_config": "src/configs/base_model.json",
  "detectron2_model_cfg": "src/configs/detectron2_configs/R-50-grid.yaml",
  "e2e_weights_path": "/pretrain/clipbert_image_text_pretrained.pt",
  "detectron2_weights_path": null,
  "bert_weights_path": null,
  "tokenizer_dir": "/pretrain/bert-base-uncased/",
  "output_dir": "/storage/anet_ret_default",
  "train_batch_size": 16,
  "val_batch_size": 16,
  "gradient_accumulation_steps": 2,
  "num_train_epochs": 80,
  "min_valid_steps": 100,
  "num_valid": 200,
  "learning_rate": 5e-5,
  "weight_decay": 1e-3,
  "decay": "linear",
  "optim": "adamw",
  "betas": [0.9, 0.98],
  "dropout": 0.1,
  "grad_norm": 5.0,
  "cnn_learning_rate": 5e-5,
  "cnn_weight_decay": 1e-3,
  "cnn_lr_decay": "linear",
  "seed":42,
  "fp16": 1,
  "loss_type": "ce",
  "itm_neg_size": 1
}