Document GTSFM Studio v0.0.2

GTSfM_Studio_Patch_Notes.md

@@ -0,0 +1,36 @@
+GTSFM Studio v0.0.2 Patch Notes
+
+- Added gtsfm run, which launches a browser-based workspace for configuring and running reconstructions.
+- Added built-in GTSFM GitHub samples, and automatic dataset-format detection.
+- Added setup checks for dependencies and optional components.
+- Added preset-aware Hydra controls that update with the selected model, plus JSON configuration import with a generated template and validation.
+
+![GTSFM Studio Patch Demo](assets/gtsfm_studio_v0.0.2_hydra_config_patch.gif)
+
+| Interface control | Resulting mapping | Relationship |
+|---|---|---|
+| Reconstruction model | `--config_name vggt` → `hydra.compose(config_name="vggt")` | Direct preset selection |
+| Generated model parameter | Exact dotted key, e.g. `cluster_optimizer.geometry_transformer.config.confidence_threshold=6.5` | One-to-one |
+| Gaussian preset | `--gaussian_splatting_config_name base_gs` → separate Hydra composition | Direct preset selection |
+| Generated Gaussian parameter | `opacity_reg=0.01` → `gaussian_splatting_optimizer.cfg.opacity_reg=0.01` | Semantically one-to-one, but the backend adds the path prefix |
+| Training steps | `gs_max_steps=7000` → `gaussian_splatting_optimizer.cfg.max_steps=7000` | Friendly alias |
+| Share camera intrinsics | Toggle → `cluster_optimizer.multiview_optimizer.bundle_adjustment_module.shared_calib=True` | Friendly alias |
+| Dataset format: Auto | Examines the directory and selects a loader plus loader overrides | Interpreted |
+| Expert Hydra overrides | Passed through unchanged after Hydra validates them | Fully one-to-one |
+| JSON configuration | Converted into model, Gaussian, and expert override sections | Interface schema, not a raw Hydra config file |
+| Hardware, Modal VM, preview interval | Process/environment/workspace settings | Not Hydra |
+
+The JSON prompt also performs interpretation. It accepts:
+```json
+{
+  "model": {
+    "cluster_optimizer.geometry_transformer.config.confidence_threshold": 6.5
+  },
+  "gaussian": {
+    "means_lr": 0.00016
+  },
+  "expert": [
+    "some.exact.hydra.key=value"
+  ]
+}
+```