MCPcopy Create free account
hub / github.com/DSL-Lab/StreamSplat / forward

Method forward

encoders/dinov2_wrapper.py:53–68  ·  view source on GitHub ↗
(self, image: torch.Tensor, mod: torch.Tensor = None)

Source from the content-addressed store, hash-verified

51
52 # @torch.compile
53 def forward(self, image: torch.Tensor, mod: torch.Tensor = None):
54 # image: [N, C, H, W]
55 # mod: [N, D] or None
56 # RGB image with [0,1] scale and properly sized
57 if self.modulation_dim is None:
58 assert mod is None, "Unexpected modulation input in dinov2 forward."
59 outs = self.model(image, is_training=True)
60 else:
61 assert mod is not None, "Modulation input is required in modulated dinov2 forward."
62 outs = self.model(image, mod=mod, is_training=True)
63 # ret = torch.cat([
64 # outs["x_norm_clstoken"].unsqueeze(dim=1),
65 # outs["x_norm_patchtokens"],
66 # ], dim=1)
67 # return ret
68 return outs["x_norm_clstoken"].unsqueeze(dim=1), outs["x_norm_patchtokens"]

Callers

nothing calls this directly

Calls

no outgoing calls

Tested by

no test coverage detected