multimodal/examples/mdetr/utils/args_parse.py

# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the BSD-style license found in the
# LICENSE file in the root directory of this source tree.

import argparse


def get_args_parser():
    parser = argparse.ArgumentParser("MDETR", add_help=False)
    parser.add_argument("--dataset_config", default=None, required=True)
    # Transformer
    parser.add_argument(
        "--enc_layers",
        default=6,
        type=int,
        help="Number of encoding layers in the transformer",
    )
    parser.add_argument(
        "--dec_layers",
        default=6,
        type=int,
        help="Number of decoding layers in the transformer",
    )
    parser.add_argument(
        "--dim_feedforward",
        default=2048,
        type=int,
        help="Intermediate size of the feedforward layers in the transformer blocks",
    )
    parser.add_argument(
        "--hidden_dim",
        default=256,
        type=int,
        help="Size of the embeddings (dimension of the transformer)",
    )
    parser.add_argument(
        "--dropout", default=0.1, type=float, help="Dropout applied in the transformer"
    )
    parser.add_argument(
        "--nheads",
        default=8,
        type=int,
        help="Number of attention heads inside the transformer's attentions",
    )
    parser.add_argument(
        "--num_classes", default=255, type=int, help="Number of classes"
    )
    parser.add_argument(
        "--num_queries", default=100, type=int, help="Number of query slots"
    )
    parser.add_argument("--pre_norm", action="store_true")

    parser.add_argument("--output_dir", default="test")
    parser.add_argument(
        "--freeze_text_encoder",
        action="store_true",
        help="Whether to freeze the weights of the text encoder",
    )
    parser.add_argument("--tokenizer_type", type=str, default="roberta-base")
    parser.add_argument("--seed", default=42, type=int)
    parser.add_argument(
        "--test",
        action="store_true",
        help="Whether to run evaluation on val or test set",
    )
    parser.add_argument(
        "--backbone",
        default="resnet101",
        type=str,
        help="Name of the convolutional backbone to use such as resnet50 resnet101 timm_tf_efficientnet_b3_ns",
    )

    parser.add_argument(
        "--no_contrastive_align_loss",
        dest="contrastive_align_loss",
        action="store_false",
        help="Whether to add contrastive alignment loss",
    )

    parser.add_argument(
        "--contrastive_loss_hdim",
        type=int,
        default=64,
        help="Projection head output size before computing the contrastive alignment loss",
    )
    parser.add_argument(
        "--temperature",
        type=float,
        default=0.07,
        help="Temperature in the contrastive alignment loss",
    )
    # Matcher
    parser.add_argument(
        "--matcher_cost_class",
        default=1,
        type=float,
        help="Class coefficient in the matching cost",
    )
    parser.add_argument(
        "--matcher_cost_bbox",
        default=5,
        type=float,
        help="L1 box coefficient in the matching cost",
    )
    parser.add_argument(
        "--matcher_cost_giou",
        default=2,
        type=float,
        help="giou box coefficient in the matching cost",
    )
    # Loss coefficients
    parser.add_argument("--ce_loss_coef", default=1, type=float)
    parser.add_argument("--bbox_loss_coef", default=5, type=float)
    parser.add_argument("--giou_loss_coef", default=2, type=float)
    parser.add_argument("--qa_loss_coef", default=1, type=float)
    parser.add_argument(
        "--no_object_weight",
        default=0.1,
        type=float,
        help="Relative classification weight of the no-object class",
    )
    parser.add_argument("--contrastive_align_loss_coef", default=1, type=float)
    parser.add_argument("--lr", default=1e-4, type=float)
    parser.add_argument("--lr_backbone", default=1e-5, type=float)
    parser.add_argument("--text_encoder_lr", default=5e-5, type=float)
    parser.add_argument("--batch_size", default=2, type=int)
    parser.add_argument("--weight_decay", default=1e-4, type=float)
    parser.add_argument("--epochs", default=40, type=int)
    parser.add_argument("--lr_drop", default=35, type=int)
    parser.add_argument(
        "--epoch_chunks",
        default=-1,
        type=int,
        help="If greater than 0, will split the training set into chunks and validate/checkpoint after each chunk",
    )
    parser.add_argument("--optimizer", default="adam", type=str)
    parser.add_argument(
        "--clip_max_norm", default=0.1, type=float, help="gradient clipping max norm"
    )
    parser.add_argument(
        "--eval_skip",
        default=1,
        type=int,
        help='do evaluation every "eval_skip" frames',
    )
    parser.add_argument("--resume", default="", help="resume from checkpoint")
    parser.add_argument("--load", default="", help="resume from checkpoint")
    parser.add_argument(
        "--start-epoch", default=0, type=int, metavar="N", help="start epoch"
    )
    parser.add_argument("--eval", action="store_true", help="Only run evaluation")
    parser.add_argument(
        "--schedule",
        default="linear_with_warmup",
        type=str,
        choices=("step", "multistep", "linear_with_warmup", "all_linear_with_warmup"),
    )
    parser.add_argument("--ema", action="store_true")
    parser.add_argument("--ema_decay", type=float, default=0.9998)
    parser.add_argument(
        "--fraction_warmup_steps",
        default=0.01,
        type=float,
        help="Fraction of total number of steps",
    )
    parser.add_argument(
        "--device", default="cuda", help="device to use for training / testing"
    )
    parser.add_argument(
        "--world-size", default=1, type=int, help="number of distributed processes"
    )
    parser.add_argument(
        "--dist-url", default="env://", help="url used to set up distributed training"
    )

    return parser