Restore detect_objects to single function like working commit 619881b

Restore .then() chain for detect_btn.click like in working commit 619881b

Add wrapper to debug button click before GPU allocation

Add comprehensive debug logging throughout the app

Debug: add log to detect_objects wrapper

Debug: add logging to update_gallery_on_upload

Disable SSR via env var GRADIO_SSR_MODE=false

Debug: log detect_btn.click setup

Disable demo.load - test manually instead

Disable SSR mode to fix demo.load

Debug: test if demo.load works at all

Add auto-run example on startup for debugging

Fix: @spaces.GPU must be on module-level function, not nested/local function

Fix: use wrapper for detect_objects to avoid .then() issues with @spaces.GPU

Add flush to detect_objects debug logging

perf: use prebuilt wheels for detectron2 and MSDeformAttn

feat: add per-example threshold support

feat: add workspace demo video example with laptop/keyboard/mouse/cup labels

debug: add logging to detect_objects entry point

fix: validate target_dir in detect_objects before processing

Reduce GPU duration to 120s to be more efficient with Zero GPU quota

Fix Zero GPU: add @spaces.GPU(duration=300) to detect_objects, remove from run_model to avoid nesting

Fix Zero GPU nested decorator issue - remove @spaces.GPU from reconstruct() and detect_objects() since they call run_model() which already has GPU decorator

Use prebuilt pytorch3d wheel instead of compiling from source

Add Dockerfile for prebuilt pytorch3d and detectron2

Fix video frame extraction and Zero GPU error handling

Improve confidence threshold UX with detection statistics

Add gr.Examples component to UI for demo video

fix: limit video frame extraction to MAX_IMAGES for Zero GPU memory

fix: repair broken torch.nn imports in vggt modules

fix: install setuptools before mmcv with --no-build-isolation

fix: add setuptools for mmcv build compatibility

fix: restore working mvp.py and add test example video

Fix MSDeformAttn: update deprecated PyTorch API for PyTorch 2.x compatibility

refactor: modularize feature extraction

fix(memory): reduce default batch size to prevent CUDA OOM on shared GPU

fix: correct attention mask dimensions

feat: add CropFormer configuration

test: validate geometry transforms

docs: add docstring to reconstruct function

fix: resolve CUDA memory fragmentation

feat: add image batch preprocessing

Fix MSDeformAttn: update deprecated PyTorch API for PyTorch 2.x compatibility

style: remove inline arch comment

chore: update numpy version constraint

fix(build): set TORCH_CUDA_ARCH_LIST for sm_86 kernel compilation

fix: handle invalid depth values

refactor(cuda): revert verbose assertion comments

fix(cuda): patch ms_deform_attn_cuda.cu for PyTorch 2.8 ABI compatibility

feat: integrate Metric3D depth estimation