summaryrefslogtreecommitdiff
path: root/worldalign/vg_extract_vision.py
diff options
context:
space:
mode:
authorYuren Hao <blackhao0426@gmail.com>2026-08-01 14:11:25 -0500
committerYuren Hao <blackhao0426@gmail.com>2026-08-01 14:11:25 -0500
commitebc2e7f5129b81db745ad15e5b78b69a95b9837c (patch)
tree07ae6afc906f0c7090d659476504f93454166357 /worldalign/vg_extract_vision.py
parenta62cf4d2a99b4a7985c61b2a7feb92a82a8218b7 (diff)
Augmentation orbits do not substitute for multiple photographs
Four random resized crops per Visual Genome image, segmented independently with their fields averaged, leave the correlation at 0.6559. Closed-world re-renders resample layout, which is nuisance by construction, so averaging removes modality-private variation; random crops perturb framing that self-supervised patch features already absorb, leaving nothing to cancel. Co-Authored-By: Claude <noreply@anthropic.com>
Diffstat (limited to 'worldalign/vg_extract_vision.py')
0 files changed, 0 insertions, 0 deletions