Add torchvision (not in the ZeroGPU base image)

depth_anything_3.utils.io.input_processor imports torchvision.transforms, but the
ZeroGPU base image ships only torch (2.11.0, CUDA 13), not torchvision. Pin
torchvision==0.26.0, which requires exactly torch==2.11.0 (verified on PyPI), so
it installs onto the existing runtime torch without pulling a different CUDA build.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

requirements.txt

@@ -1,7 +1,11 @@
 # FaceAnything Space dependencies.
 #
-# torch / torchvision are provided by the Space runtime (ZeroGPU / GPU image) and
-# are intentionally NOT pinned here so they match the platform CUDA build.
+# torch is provided by the Space runtime (ZeroGPU image ships torch==2.11.0, CUDA
+# 13) so it is NOT pinned here. torchvision, however, is NOT in the base image and
+# must be installed — pinned to the build that matches the runtime torch
+# (torchvision 0.26.0 requires torch==2.11.0), so it slots onto the existing torch
+# without dragging in a different CUDA build. Bump both together if HF changes the
+# base image torch version (check the build logs).
 #
 # xformers is intentionally omitted: the model falls back to a pure-PyTorch SwiGLU
 # when it is absent (see dinov2/layers/swiglu_ffn.py), and pinning it tends to
@@ -9,6 +13,7 @@
 # kernel and your torch build is compatible.
 
 spaces
+torchvision==0.26.0    # matches runtime torch==2.11.0 (NOT in the base image)
 
 numpy==1.26.4
 scipy==1.16.3