File size: 7,305 Bytes
ac239ba |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 |
# 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 logging
import torch
from hydra import compose
from hydra.utils import instantiate
from omegaconf import OmegaConf
HF_MODEL_ID_TO_FILENAMES = {
"facebook/sam2-hiera-tiny": (
"configs/sam2/sam2_hiera_t.yaml",
"sam2_hiera_tiny.pt",
),
"facebook/sam2-hiera-small": (
"configs/sam2/sam2_hiera_s.yaml",
"sam2_hiera_small.pt",
),
"facebook/sam2-hiera-base-plus": (
"configs/sam2/sam2_hiera_b+.yaml",
"sam2_hiera_base_plus.pt",
),
"facebook/sam2-hiera-large": (
"configs/sam2/sam2_hiera_l.yaml",
"sam2_hiera_large.pt",
),
"facebook/sam2.1-hiera-tiny": (
"configs/sam2.1/sam2.1_hiera_t.yaml",
"sam2.1_hiera_tiny.pt",
),
"facebook/sam2.1-hiera-small": (
"configs/sam2.1/sam2.1_hiera_s.yaml",
"sam2.1_hiera_small.pt",
),
"facebook/sam2.1-hiera-base-plus": (
"configs/sam2.1/sam2.1_hiera_b+.yaml",
"sam2.1_hiera_base_plus.pt",
),
"facebook/sam2.1-hiera-large": (
"configs/sam2.1/sam2.1_hiera_l.yaml",
"sam2.1_hiera_large.pt",
),
}
def get_best_available_device():
"""
Get the best available device in the order: CUDA, MPS, CPU
Returns: device string for torch.device
"""
if torch.cuda.is_available():
return "cuda"
elif hasattr(torch.backends, 'mps') and torch.backends.mps.is_available():
return "mps"
else:
return "cpu"
def build_sam2(
config_file,
ckpt_path=None,
device=None,
mode="eval",
hydra_overrides_extra=[],
apply_postprocessing=True,
**kwargs,
):
# Use the provided device or get the best available one
device = device or get_best_available_device()
logging.info(f"Using device: {device}")
if apply_postprocessing:
hydra_overrides_extra = hydra_overrides_extra.copy()
hydra_overrides_extra += [
# dynamically fall back to multi-mask if the single mask is not stable
"++model.sam_mask_decoder_extra_args.dynamic_multimask_via_stability=true",
"++model.sam_mask_decoder_extra_args.dynamic_multimask_stability_delta=0.05",
"++model.sam_mask_decoder_extra_args.dynamic_multimask_stability_thresh=0.98",
]
# Read config and init model
cfg = compose(config_name=config_file, overrides=hydra_overrides_extra)
OmegaConf.resolve(cfg)
model = instantiate(cfg.model, _recursive_=True)
_load_checkpoint(model, ckpt_path)
model = model.to(device)
if mode == "eval":
model.eval()
return model
def build_sam2_video_predictor(
config_file,
ckpt_path=None,
device=None,
mode="eval",
hydra_overrides_extra=[],
apply_postprocessing=True,
**kwargs,
):
# Use the provided device or get the best available one
device = device or get_best_available_device()
logging.info(f"Using device: {device}")
hydra_overrides = [
"++model._target_=sam2.sam2_video_predictor.SAM2VideoPredictor",
]
if apply_postprocessing:
hydra_overrides_extra = hydra_overrides_extra.copy()
hydra_overrides_extra += [
# dynamically fall back to multi-mask if the single mask is not stable
"++model.sam_mask_decoder_extra_args.dynamic_multimask_via_stability=true",
"++model.sam_mask_decoder_extra_args.dynamic_multimask_stability_delta=0.05",
"++model.sam_mask_decoder_extra_args.dynamic_multimask_stability_thresh=0.98",
# the sigmoid mask logits on interacted frames with clicks in the memory encoder so that the encoded masks are exactly as what users see from clicking
"++model.binarize_mask_from_pts_for_mem_enc=true",
# fill small holes in the low-res masks up to `fill_hole_area` (before resizing them to the original video resolution)
"++model.fill_hole_area=8",
]
hydra_overrides.extend(hydra_overrides_extra)
# Read config and init model
cfg = compose(config_name=config_file, overrides=hydra_overrides)
OmegaConf.resolve(cfg)
model = instantiate(cfg.model, _recursive_=True)
_load_checkpoint(model, ckpt_path)
model = model.to(device)
if mode == "eval":
model.eval()
return model
def build_sam2_video_predictor_npz(
config_file,
ckpt_path=None,
device=None,
mode="eval",
hydra_overrides_extra=[],
apply_postprocessing=True,
**kwargs,
):
# Use the provided device or get the best available one
device = device or get_best_available_device()
logging.info(f"Using device: {device}")
hydra_overrides = [
"++model._target_=sam2.sam2_video_predictor_npz.SAM2VideoPredictorNPZ",
]
if apply_postprocessing:
hydra_overrides_extra = hydra_overrides_extra.copy()
hydra_overrides_extra += [
# dynamically fall back to multi-mask if the single mask is not stable
"++model.sam_mask_decoder_extra_args.dynamic_multimask_via_stability=true",
"++model.sam_mask_decoder_extra_args.dynamic_multimask_stability_delta=0.05",
"++model.sam_mask_decoder_extra_args.dynamic_multimask_stability_thresh=0.98",
# the sigmoid mask logits on interacted frames with clicks in the memory encoder so that the encoded masks are exactly as what users see from clicking
"++model.binarize_mask_from_pts_for_mem_enc=true",
# fill small holes in the low-res masks up to `fill_hole_area` (before resizing them to the original video resolution)
"++model.fill_hole_area=8",
]
hydra_overrides.extend(hydra_overrides_extra)
# Read config and init model
cfg = compose(config_name=config_file, overrides=hydra_overrides)
OmegaConf.resolve(cfg)
model = instantiate(cfg.model, _recursive_=True)
_load_checkpoint(model, ckpt_path)
model = model.to(device)
if mode == "eval":
model.eval()
return model
def _hf_download(model_id):
from huggingface_hub import hf_hub_download
config_name, checkpoint_name = HF_MODEL_ID_TO_FILENAMES[model_id]
ckpt_path = hf_hub_download(repo_id=model_id, filename=checkpoint_name)
return config_name, ckpt_path
def build_sam2_hf(model_id, **kwargs):
config_name, ckpt_path = _hf_download(model_id)
return build_sam2(config_file=config_name, ckpt_path=ckpt_path, **kwargs)
def build_sam2_video_predictor_hf(model_id, **kwargs):
config_name, ckpt_path = _hf_download(model_id)
return build_sam2_video_predictor(
config_file=config_name, ckpt_path=ckpt_path, **kwargs
)
def _load_checkpoint(model, ckpt_path):
if ckpt_path is not None:
sd = torch.load(ckpt_path, map_location="cpu", weights_only=True)["model"]
missing_keys, unexpected_keys = model.load_state_dict(sd)
if missing_keys:
logging.error(missing_keys)
raise RuntimeError()
if unexpected_keys:
logging.error(unexpected_keys)
raise RuntimeError()
logging.info("Loaded checkpoint sucessfully")
|