euler-eval 2.0.1__tar.gz → 2.2.0__tar.gz

{euler_eval-2.0.1 → euler_eval-2.2.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: euler-eval
-Version: 2.0.1
+Version: 2.2.0
 Summary: Depth map evaluation toolkit with comprehensive metrics
 Author: Depth Eval Contributors
 License: MIT
@@ -22,6 +22,7 @@ Requires-Dist: Pillow>=8.0.0
 Requires-Dist: torch>=1.9.0
 Requires-Dist: torchvision>=0.10.0
 Requires-Dist: lpips>=0.1.4
+Requires-Dist: torchmetrics>=1.0
 Requires-Dist: tqdm>=4.62.0
 Requires-Dist: euler-loading
 Requires-Dist: ds-crawler
@@ -294,6 +295,7 @@ Controls sanity check thresholds. See [metrics_config.json](metrics_config.json)
 | LPIPS | `depth.image_quality.lpips` | Learned Perceptual Image Patch Similarity |
 | FID | `depth.image_quality.fid` | Fréchet Inception Distance (dataset-level distribution metric) |
 | KID | `depth.image_quality.kid_mean`, `kid_std` | Kernel Inception Distance (mean and std) |
+| Standard depth metrics | `depth.standard.{image_mean,image_median,pixel_pool}.*` | Monocular-depth metrics with explicit reducers: `absrel`, `sqrel`, `mae`, `rmse`, `rmse_log`, `log10`, `silog`, `delta1`, `delta2`, `delta3` |
 | AbsRel | `depth.depth_metrics.absrel` | Absolute Relative Error (\|pred-gt\|/gt), reported as median and p90 |
 | RMSE | `depth.depth_metrics.rmse` | Root Mean Square Error, reported as median and p90 |
 | SILog | `depth.depth_metrics.silog` | Scale-Invariant Log Error, reported as mean, median, and p90 |
@@ -375,6 +377,8 @@ For depth outputs:
 - `depth_raw`: metric-space depth without any post-hoc alignment.
 - `depth_aligned`: metric-space depth after configured alignment mode.
 - `depth`: backward-compatible alias of `depth_aligned`.
+- `standard`: explicit monocular-depth metrics with three reducers:
+  `image_mean`, `image_median`, and `pixel_pool`.
 
 Previous single-depth structure (kept under `depth`) is:
 
@@ -389,6 +393,44 @@ Previous single-depth structure (kept under `depth`) is:
       "kid_mean": 0.005,
       "kid_std": 0.002
     },
+    "standard": {
+      "image_mean": {
+        "absrel": 0.08,
+        "sqrel": 0.04,
+        "mae": 0.62,
+        "rmse": 1.20,
+        "rmse_log": 0.11,
+        "log10": 0.04,
+        "silog": 0.08,
+        "delta1": 0.91,
+        "delta2": 0.97,
+        "delta3": 0.99
+      },
+      "image_median": {
+        "absrel": 0.07,
+        "sqrel": 0.03,
+        "mae": 0.58,
+        "rmse": 1.10,
+        "rmse_log": 0.10,
+        "log10": 0.04,
+        "silog": 0.07,
+        "delta1": 0.92,
+        "delta2": 0.98,
+        "delta3": 0.99
+      },
+      "pixel_pool": {
+        "absrel": 0.08,
+        "sqrel": 0.04,
+        "mae": 0.61,
+        "rmse": 1.18,
+        "rmse_log": 0.11,
+        "log10": 0.04,
+        "silog": 0.08,
+        "delta1": 0.91,
+        "delta2": 0.97,
+        "delta3": 0.99
+      }
+    },
     "depth_metrics": {
       "absrel": { "median": 0.05, "p90": 0.12 },
       "rmse":   { "median": 1.2, "p90": 3.1 },

{euler_eval-2.0.1 → euler_eval-2.2.0}/README.md

@@ -257,6 +257,7 @@ Controls sanity check thresholds. See [metrics_config.json](metrics_config.json)
 | LPIPS | `depth.image_quality.lpips` | Learned Perceptual Image Patch Similarity |
 | FID | `depth.image_quality.fid` | Fréchet Inception Distance (dataset-level distribution metric) |
 | KID | `depth.image_quality.kid_mean`, `kid_std` | Kernel Inception Distance (mean and std) |
+| Standard depth metrics | `depth.standard.{image_mean,image_median,pixel_pool}.*` | Monocular-depth metrics with explicit reducers: `absrel`, `sqrel`, `mae`, `rmse`, `rmse_log`, `log10`, `silog`, `delta1`, `delta2`, `delta3` |
 | AbsRel | `depth.depth_metrics.absrel` | Absolute Relative Error (\|pred-gt\|/gt), reported as median and p90 |
 | RMSE | `depth.depth_metrics.rmse` | Root Mean Square Error, reported as median and p90 |
 | SILog | `depth.depth_metrics.silog` | Scale-Invariant Log Error, reported as mean, median, and p90 |
@@ -338,6 +339,8 @@ For depth outputs:
 - `depth_raw`: metric-space depth without any post-hoc alignment.
 - `depth_aligned`: metric-space depth after configured alignment mode.
 - `depth`: backward-compatible alias of `depth_aligned`.
+- `standard`: explicit monocular-depth metrics with three reducers:
+  `image_mean`, `image_median`, and `pixel_pool`.
 
 Previous single-depth structure (kept under `depth`) is:
 
@@ -352,6 +355,44 @@ Previous single-depth structure (kept under `depth`) is:
       "kid_mean": 0.005,
       "kid_std": 0.002
     },
+    "standard": {
+      "image_mean": {
+        "absrel": 0.08,
+        "sqrel": 0.04,
+        "mae": 0.62,
+        "rmse": 1.20,
+        "rmse_log": 0.11,
+        "log10": 0.04,
+        "silog": 0.08,
+        "delta1": 0.91,
+        "delta2": 0.97,
+        "delta3": 0.99
+      },
+      "image_median": {
+        "absrel": 0.07,
+        "sqrel": 0.03,
+        "mae": 0.58,
+        "rmse": 1.10,
+        "rmse_log": 0.10,
+        "log10": 0.04,
+        "silog": 0.07,
+        "delta1": 0.92,
+        "delta2": 0.98,
+        "delta3": 0.99
+      },
+      "pixel_pool": {
+        "absrel": 0.08,
+        "sqrel": 0.04,
+        "mae": 0.61,
+        "rmse": 1.18,
+        "rmse_log": 0.11,
+        "log10": 0.04,
+        "silog": 0.08,
+        "delta1": 0.91,
+        "delta2": 0.97,
+        "delta3": 0.99
+      }
+    },
     "depth_metrics": {
       "absrel": { "median": 0.05, "p90": 0.12 },
       "rmse":   { "median": 1.2, "p90": 3.1 },

{euler_eval-2.0.1 → euler_eval-2.2.0}/euler_eval/cli.py

@@ -65,13 +65,20 @@ _DEPTH_ALIGNMENT_AXIS = AxisDeclaration(
 
 _DEPTH_CATEGORY_AXIS = AxisDeclaration(
     position=1,
-    values=("image_quality", "depth_metrics", "geometric_metrics"),
+    values=("image_quality", "standard", "depth_metrics", "geometric_metrics"),
     optional=True,
     description="Metric category",
 )
 
-_BENCHMARK_BIN_AXIS = AxisDeclaration(
+_DEPTH_REDUCTION_AXIS = AxisDeclaration(
     position=2,
+    values=("image_mean", "image_median", "pixel_pool"),
+    optional=True,
+    description="Dataset reduction mode",
+)
+
+_BENCHMARK_BIN_AXIS = AxisDeclaration(
+    position=3,
     values=("all", "near", "mid", "far"),
     optional=True,
     description="Depth benchmark bin",
@@ -99,7 +106,11 @@ _RGB_BENCHMARK_BIN_AXIS = AxisDeclaration(
 
 
 def _depth_eval_axes(*, benchmark: bool = False) -> dict[str, AxisDeclaration]:
-    axes = {"alignment": _DEPTH_ALIGNMENT_AXIS, "category": _DEPTH_CATEGORY_AXIS}
+    axes = {
+        "alignment": _DEPTH_ALIGNMENT_AXIS,
+        "category": _DEPTH_CATEGORY_AXIS,
+        "reduction": _DEPTH_REDUCTION_AXIS,
+    }
     if benchmark:
         axes["bin"] = _BENCHMARK_BIN_AXIS
     return axes
@@ -126,6 +137,46 @@ _DEPTH_EVAL_DESCRIPTIONS = {
     "lpips": MetricDescription(is_higher_better=False, display_name="LPIPS"),
     "fid": MetricDescription(is_higher_better=False, display_name="FID"),
     "kid_mean": MetricDescription(is_higher_better=False, display_name="KID Mean"),
+    "kid_std": MetricDescription(is_higher_better=False, display_name="KID Std"),
+    "absrel": MetricDescription(
+        is_higher_better=False, display_name="AbsRel"
+    ),
+    "sqrel": MetricDescription(
+        is_higher_better=False, display_name="SqRel"
+    ),
+    "mae": MetricDescription(
+        is_higher_better=False, unit="meters", display_name="MAE"
+    ),
+    "rmse": MetricDescription(
+        is_higher_better=False, unit="meters", display_name="RMSE"
+    ),
+    "rmse_log": MetricDescription(
+        is_higher_better=False, scale="log", display_name="RMSE Log"
+    ),
+    "log10": MetricDescription(
+        is_higher_better=False, scale="log", display_name="Log10 Error"
+    ),
+    "silog": MetricDescription(
+        is_higher_better=False, scale="log", display_name="SILog"
+    ),
+    "delta1": MetricDescription(
+        is_higher_better=True,
+        min_value=0.0,
+        max_value=1.0,
+        display_name="δ < 1.25",
+    ),
+    "delta2": MetricDescription(
+        is_higher_better=True,
+        min_value=0.0,
+        max_value=1.0,
+        display_name="δ < 1.25²",
+    ),
+    "delta3": MetricDescription(
+        is_higher_better=True,
+        min_value=0.0,
+        max_value=1.0,
+        display_name="δ < 1.25³",
+    ),
     "absrel.median": MetricDescription(
         is_higher_better=False, display_name="AbsRel (Median)"
     ),
@@ -869,6 +920,7 @@ def main():
 
             # Inject benchmark bin metrics under the existing category
             # keys so that the bin axis decomposes correctly:
+            #   depth.eval.aligned.standard.image_mean.{bin}.absrel
             #   depth.eval.aligned.depth_metrics.{bin}.absrel.median
             #   depth.eval.aligned.geometric_metrics.{bin}.normal_consistency.mean_angle
             depth_benchmark = depth_results.get("depth_benchmark")
@@ -879,7 +931,12 @@ def main():
                     for category, metrics in bin_summary.items():
                         cleaned = _clean_metric_tree(metrics)
                         if cleaned:
-                            aligned.setdefault(category, {})[bn] = cleaned
+                            if category == "standard":
+                                bucket = aligned.setdefault(category, {})
+                                for reduction, reduction_metrics in cleaned.items():
+                                    bucket.setdefault(reduction, {})[bn] = reduction_metrics
+                            else:
+                                aligned.setdefault(category, {})[bn] = cleaned
                 depth_save["metricSet"]["metadata"]["benchmark"] = {
                     "depth_range": depth_benchmark["boundaries"]["range"],
                     "boundaries": depth_benchmark["boundaries"],
@@ -1084,6 +1141,7 @@ def main():
                 fov_domain=rays_meta["fov_domain"],
                 gt_name=gt.get("name", "GT"),
                 pred_name=ds_name,
+                num_workers=args.num_workers,
                 verbose=args.verbose,
                 sanity_checker=sanity_checker,
             )