EfficientSAM/EdgeSAM/setup_edge_sam.py

# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.

# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.

import torch

from functools import partial

from segment_anything.modeling import ImageEncoderViT, MaskDecoder, PromptEncoder, Sam, TwoWayTransformer
from EdgeSAM.rep_vit import RepViT


prompt_embed_dim = 256
image_size = 1024
vit_patch_size = 16
image_embedding_size = image_size // vit_patch_size


def build_edge_sam(checkpoint=None, upsample_mode="bicubic"):
    image_encoder = RepViT(
        arch="m1",
        img_size=image_size,
        upsample_mode=upsample_mode
    )
    return _build_sam(image_encoder, checkpoint)


sam_model_registry = {
    "default": build_edge_sam,
    "edge_sam": build_edge_sam,
}

def _build_sam_encoder(
    encoder_embed_dim,
    encoder_depth,
    encoder_num_heads,
    encoder_global_attn_indexes,
):
    image_encoder = ImageEncoderViT(
        depth=encoder_depth,
        embed_dim=encoder_embed_dim,
        img_size=image_size,
        mlp_ratio=4,
        norm_layer=partial(torch.nn.LayerNorm, eps=1e-6),
        num_heads=encoder_num_heads,
        patch_size=vit_patch_size,
        qkv_bias=True,
        use_rel_pos=True,
        global_attn_indexes=encoder_global_attn_indexes,
        window_size=14,
        out_chans=prompt_embed_dim,
    )
    return image_encoder


def _build_sam(
    image_encoder,
    checkpoint=None,
):
    sam = Sam(
        image_encoder=image_encoder,
        prompt_encoder=PromptEncoder(
            embed_dim=prompt_embed_dim,
            image_embedding_size=(image_embedding_size, image_embedding_size),
            input_image_size=(image_size, image_size),
            mask_in_chans=16,
        ),
        mask_decoder=MaskDecoder(
            num_multimask_outputs=3,
            transformer=TwoWayTransformer(
                depth=2,
                embedding_dim=prompt_embed_dim,
                mlp_dim=2048,
                num_heads=8,
            ),
            transformer_dim=prompt_embed_dim,
            iou_head_depth=3,
            iou_head_hidden_dim=256,
        ),
        pixel_mean=[123.675, 116.28, 103.53],
        pixel_std=[58.395, 57.12, 57.375],
    )
    sam.eval()
    if checkpoint is not None:
        with open(checkpoint, "rb") as f:
            state_dict = torch.load(f, map_location="cpu")
        sam.load_state_dict(state_dict)
    return sam