Disable mask caching and adjust defaults

ARCHITECTURE.md

@@ -5,7 +5,7 @@ This document describes the flow in the current Gradio app (`app/ui.py`), from i
 ## Data and Models
 - **Inputs**: Images from `data/Image/VISLOC` (populated via `list_all_data_inputs`) with a 5% border crop (`crop_nonblack`) to drop black padding. Supported extensions: jpg/jpeg/png (any case).
 - **Depth model**: Depth Anything 3, cached per model id (`DepthEngine`). Inference caps the long side to `process_res_cap` (default 1024) using `upper_bound_resize` before predicting.
-- **Segmentation model**: SAM3 (`facebook/sam3`) for promptable water/road masking. Loaded once per model id; masks are cached per `(model_id, source_path, prompts, thresholds, max_side)`. Default `segmentation_max_side` is 384 to keep it fast on CUDA.
+- **Segmentation model**: SAM3 (`facebook/sam3`) for promptable water/road masking. Loaded once per model id; masks are recomputed every run (no caching). Default `segmentation_max_side` is 384 to keep it fast on CUDA.
 
 ## Constants and Defaults
 - Altitude/FOV defaults: 450 m, 90°.
@@ -56,7 +56,7 @@ This document describes the flow in the current Gradio app (`app/ui.py`), from i
 
 ## Caching and State
 - Depth model cache keyed by model id (`DepthEngine`).
-- SAM3 cache keyed by model id + source + prompts + thresholds + max_side (`SegmenterService`).
+- SAM3 masks are not cached; every run recomputes them to reflect real-time cost. Segmenter models stay loaded per id.
 - `images_state` holds the latest rendered layers; overlay-only changes don’t rerun inference. Prompt changes only re-trigger processing on submit/Run, not every keystroke.
 
 ## User Controls and Effects

app/config.py

@@ -28,7 +28,7 @@ class AnalyzerSettings:
     grad_thresh: float = 0.1
     clearance_factor: float = 0.0
     process_res_cap: int = 1024
-    depth_smoothing_base: float = 0.8
+    depth_smoothing_base: float = 0.0
     segmentation_max_side: int = SEGMENTATION_MAX_SIDE
     segmentation_score_thresh: float = SEGMENTATION_SCORE_THRESH
     segmentation_mask_thresh: float = SEGMENTATION_MASK_THRESH
@@ -36,7 +36,7 @@ class AnalyzerSettings:
     road_prompt: str = ROAD_PROMPT
     coverage_strictness: float = 0.95
     openness_weight: float = 0.3
-    texture_threshold: float = 0.5
+    texture_threshold: float = 0.1
     altitude_m: float = DEFAULT_ALTITUDE_M
     fov_deg: float = ASSUMED_FOV_DEG
     model_id: str = DEFAULT_MODEL_ID

app/segmentation.py

@@ -126,7 +126,6 @@ class SegmenterService:
     def __init__(self, model_id: str = SEGMENTATION_MODEL_ID):
         self.model_id = model_id
         self._segmenters: Dict[str, SemanticSegmenter] = {}
-        self._mask_cache: Dict[tuple[str, str, int], dict[str, np.ndarray]] = {}
 
     def _get_segmenter(self, model_id: str) -> SemanticSegmenter:
         if model_id not in self._segmenters:
@@ -136,36 +135,23 @@ class SegmenterService:
     def get_masks(self, request: SegmenterRequest) -> dict[str, np.ndarray]:
         if not (request.want_water or request.want_road):
             return {}
-        key = (
-            self.model_id,
-            request.source_path or "",
-            request.max_side,
-            (request.water_prompt or "").strip(),
-            (request.road_prompt or "").strip(),
-            float(request.score_threshold),
-            float(request.mask_threshold),
-        )
-        masks = self._mask_cache.get(key)
-        if masks is None:
-            segmenter = self._get_segmenter(self.model_id)
-            prompts: dict[str, str] = {}
-            if request.want_water and request.water_prompt:
-                prompts["water"] = request.water_prompt
-            if request.want_road and request.road_prompt:
-                prompts["road"] = request.road_prompt
-            try:
-                masks = segmenter.segment(
-                    request.image,
-                    request.max_side,
-                    prompts=prompts,
-                    score_threshold=float(request.score_threshold),
-                    mask_threshold=float(request.mask_threshold),
-                )
-            except RuntimeError as exc:
-                print(f"[WARN] Segmentation failed; skipping masks: {exc}")
-                masks = {}
-            if request.source_path and masks:
-                self._mask_cache[key] = masks
+        segmenter = self._get_segmenter(self.model_id)
+        prompts: dict[str, str] = {}
+        if request.want_water and request.water_prompt:
+            prompts["water"] = request.water_prompt
+        if request.want_road and request.road_prompt:
+            prompts["road"] = request.road_prompt
+        try:
+            masks = segmenter.segment(
+                request.image,
+                request.max_side,
+                prompts=prompts,
+                score_threshold=float(request.score_threshold),
+                mask_threshold=float(request.mask_threshold),
+            )
+        except RuntimeError as exc:
+            print(f"[WARN] Segmentation failed; skipping masks: {exc}")
+            masks = {}
         result: dict[str, np.ndarray] = {}
         if request.want_water and masks.get("water") is not None:
             result["water"] = masks["water"]