nodes/kijai-ComfyUI-DepthAnythingV2
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Add giant model

Browse Source 
https://huggingface.co/Nap/depth_anything_v2_vitg
This commit is contained in:
kijai
2025-06-16 16:16:50 +03:00
parent 9d7cb8c1e5
commit d505cbca99
1 changed files with 67 additions and 51 deletions
Download Patch File Download Diff File Expand all files Collapse all files
nodes.py
+67 -51
View File
@@ -32,14 +32,19 @@ class DownloadAndLoadDepthAnythingV2Model:
'depth_anything_v2_vitb_fp32.safetensors',
'depth_anything_v2_vitl_fp16.safetensors',
'depth_anything_v2_vitl_fp32.safetensors',
'depth_anything_v2_vitg_fp32.safetensors',
'depth_anything_v2_metric_hypersim_vitl_fp32.safetensors',
'depth_anything_v2_metric_vkitti_vitl_fp32.safetensors'
],
{
"default": 'depth_anything_v2_vitl_fp32.safetensors'
}),
},
}
{
"default": 'depth_anything_v2_vitl_fp32.safetensors'
}),
},
"optional": {
"precision": (["auto", "bf16", "fp16", "fp32"], {"default": "auto"},)
}
}
RETURN_TYPES = ("DAMODEL",)
RETURN_NAMES = ("da_v2_model",)
@@ -52,59 +57,70 @@ https://huggingface.co/Kijai/DepthAnythingV2-safetensors/tree/main
fp16 reduces quality by a LOT, not recommended.
"""
def loadmodel(self, model):
def loadmodel(self, model, precision="fp32"):
device = mm.get_torch_device()
dtype = torch.float16 if "fp16" in model else torch.float32
if precision == "auto":
dtype = torch.float16 if "fp16" in model else torch.float32
elif precision == "bf16":
dtype = torch.bfloat16
elif precision == "fp16":
dtype = torch.float16
elif precision == "fp32":
dtype = torch.float32
model_configs = {
'vits': {'encoder': 'vits', 'features': 64, 'out_channels': [48, 96, 192, 384]},
'vitb': {'encoder': 'vitb', 'features': 128, 'out_channels': [96, 192, 384, 768]},
'vitl': {'encoder': 'vitl', 'features': 256, 'out_channels': [256, 512, 1024, 1024]},
#'vitg': {'encoder': 'vitg', 'features': 384, 'out_channels': [1536, 1536, 1536, 1536]}
'vitg': {'encoder': 'vitg', 'features': 384, 'out_channels': [1536, 1536, 1536, 1536]}
}
custom_config = {
'model_name': model,
}
if not hasattr(self, 'model') or self.model == None or custom_config != self.current_config:
self.current_config = custom_config
download_path = os.path.join(folder_paths.models_dir, "depthanything")
model_path = os.path.join(download_path, model)
download_path = os.path.join(folder_paths.models_dir, "depthanything")
model_path = os.path.join(download_path, model)
if not os.path.exists(model_path):
print(f"Downloading model to: {model_path}")
from huggingface_hub import snapshot_download
snapshot_download(repo_id="Kijai/DepthAnythingV2-safetensors",
allow_patterns=[f"*{model}*"],
local_dir=download_path,
local_dir_use_symlinks=False)
if "vitg" in model:
repo = "Nap/depth_anything_v2_vitg"
else:
repo = "Kijai/DepthAnythingV2-safetensors"
print(f"Loading model from: {model_path}")
if not os.path.exists(model_path):
print(f"Downloading model to: {model_path}")
from huggingface_hub import snapshot_download
snapshot_download(repo_id=repo,
allow_patterns=[f"*{model}*"],
local_dir=download_path,
local_dir_use_symlinks=False)
if "vitl" in model:
encoder = "vitl"
elif "vitb" in model:
encoder = "vitb"
elif "vits" in model:
encoder = "vits"
print(f"Loading model from: {model_path}")
if "hypersim" in model:
max_depth = 20.0
if "vitg" in model:
encoder = "vitg"
elif "vitl" in model:
encoder = "vitl"
elif "vitb" in model:
encoder = "vitb"
elif "vits" in model:
encoder = "vits"
if "hypersim" in model:
max_depth = 20.0
else:
max_depth = 80.0
with (init_empty_weights() if is_accelerate_available else nullcontext()):
if 'metric' in model:
self.model = DepthAnythingV2(**{**model_configs[encoder], 'is_metric': True, 'max_depth': max_depth})
else:
max_depth = 80.0
self.model = DepthAnythingV2(**model_configs[encoder])
state_dict = load_torch_file(model_path)
if is_accelerate_available:
for key in state_dict:
set_module_tensor_to_device(self.model, key, device=device, dtype=dtype, value=state_dict[key])
else:
self.model.load_state_dict(state_dict)
with (init_empty_weights() if is_accelerate_available else nullcontext()):
if 'metric' in model:
self.model = DepthAnythingV2(**{**model_configs[encoder], 'is_metric': True, 'max_depth': max_depth})
else:
self.model = DepthAnythingV2(**model_configs[encoder])
state_dict = load_torch_file(model_path)
if is_accelerate_available:
for key in state_dict:
set_module_tensor_to_device(self.model, key, device=device, dtype=dtype, value=state_dict[key])
else:
self.model.load_state_dict(state_dict)
self.model.eval()
self.model.eval()
da_model = {
"model": self.model,
@@ -162,15 +178,15 @@ https://depth-anything-v2.github.io
depth = (depth - depth.min()) / (depth.max() - depth.min())
out.append(depth.cpu())
pbar.update(1)
model.to(offload_device)
depth_out = torch.cat(out, dim=0)
depth_out = depth_out.unsqueeze(-1).repeat(1, 1, 1, 3).cpu().float()
model.to(offload_device)
mm.soft_empty_cache()
depth_out = torch.cat(out, dim=0)
depth_out = depth_out.unsqueeze(-1).repeat(1, 1, 1, 3).cpu().float()
final_H = (orig_H // 2) * 2
final_W = (orig_W // 2) * 2
if depth_out.shape[1] != final_H or depth_out.shape[2] != final_W:
depth_out = F.interpolate(depth_out.permute(0, 3, 1, 2), size=(final_H, final_W), mode="bilinear").permute(0, 2, 3, 1)
depth_out = (depth_out - depth_out.min()) / (depth_out.max() - depth_out.min())