Upload folder using huggingface_hub

.gitattributes

@@ -149,3 +149,5 @@ exported_maps/town01_official/Town01.usd filter=lfs diff=lfs merge=lfs -text
 rendered_scenes/priority_from_right/run_000/Town05_ground.usd filter=lfs diff=lfs merge=lfs -text
 rendered_scenes/priority_from_right/run_000_rainy/Town05_ground.usd filter=lfs diff=lfs merge=lfs -text
 rendered_scenes/priority_road/run_000/Town05_ground.usd filter=lfs diff=lfs merge=lfs -text
+output/priority_from_right/run_000/Town05_ground.usd filter=lfs diff=lfs merge=lfs -text
+output/priority_from_right/run_000_rainy/Town05_ground.usd filter=lfs diff=lfs merge=lfs -text

output/priority_from_right/run_000/Town05_ground.usd

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dda7af9ab36ad9c2731e0ac808b79707ad576ef8b3a483c7c8252b334dc65891
+size 307384

output/priority_from_right/run_000/cosmos_input/README.md

@@ -0,0 +1,42 @@
+# Cosmos-Transfer2.5 inputs for `priority_from_right`
+
+This folder is auto-generated by the SceneDriveBench pipeline. It contains
+single-view inference specs for the Cosmos-Transfer2.5 base model.
+
+## Which spec should I run?
+
+**`multicontrol/`** — the recommended option. Combines depth + edge + seg
+controls with weights tuned to the CARLA SDG cookbook (depth=1.0, edge=0.2,
+seg=1.0). This produces the most photoreal output with minimal artifacts.
+
+`depth/`, `edge/`, `seg/` — single-control ablations. Useful for studying
+how each modality alone shapes the output. **Edge-only at weight 1.0
+typically produces ghostly outline-y results — avoid unless that is the
+specific signal you are studying.**
+
+## Run from the cosmos-transfer2.5 repo:
+
+```bash
+python examples/inference.py \
+  -i <path-to-this-folder>/multicontrol/priority_from_right_front_multicontrol_spec.json \
+  -o outputs/priority_from_right_front
+```
+
+Repeat for `rear`, `left`, `right` to get all four views.
+
+All paths inside the spec JSONs are relative to the spec's own folder, so
+the whole run directory is portable across machines.
+
+## If the output doesn't look right, tune these knobs in the spec JSON
+
+| Knob | What it does | Try if... |
+|---|---|---|
+| `guidance` (default 5) | How hard the prompt pulls the output | Output looks too CARLA-like / videogame-y → 6-7 |
+| `vis.control_weight` (default 0.3) | Anchor strength to the (blurred) input | Foreground objects look blurry → drop to 0.2 |
+| `vis.control_weight` (default 0.3) | Anchor strength to the (blurred) input | Sky / distant regions hallucinate trees/fractals → raise to 0.5-0.7 |
+| `seg.control_weight` (default 1.0) | Adherence to segmentation labels | Buildings / road boundaries break → keep at 1.0; raise to 1.2 if they smear |
+| `edge.control_weight` (default 0.2) | Adherence to outlines | Cars/buildings look soft → 0.3-0.4 |
+| `depth.control_weight` (default 1.0) | 3D structure constraint | Geometry looks wrong → leave at 1.0 |
+
+After editing, re-run the same `examples/inference.py` command. The output
+folder is the only thing that needs to change between runs.

output/priority_from_right/run_000/cosmos_input/depth/priority_from_right_front_depth_spec.json

@@ -0,0 +1,10 @@
+{
+    "name": "priority_from_right_front_depth",
+    "prompt_path": "../prompt.txt",
+    "video_path": "../../videos/rgb_front.mp4",
+    "guidance": 3,
+    "depth": {
+        "control_path": "../../videos/depth_front.mp4",
+        "control_weight": 1.0
+    }
+}

output/priority_from_right/run_000/cosmos_input/depth/priority_from_right_left_depth_spec.json

@@ -0,0 +1,10 @@
+{
+    "name": "priority_from_right_left_depth",
+    "prompt_path": "../prompt.txt",
+    "video_path": "../../videos/rgb_left.mp4",
+    "guidance": 3,
+    "depth": {
+        "control_path": "../../videos/depth_left.mp4",
+        "control_weight": 1.0
+    }
+}

output/priority_from_right/run_000/cosmos_input/depth/priority_from_right_rear_depth_spec.json

@@ -0,0 +1,10 @@
+{
+    "name": "priority_from_right_rear_depth",
+    "prompt_path": "../prompt.txt",
+    "video_path": "../../videos/rgb_rear.mp4",
+    "guidance": 3,
+    "depth": {
+        "control_path": "../../videos/depth_rear.mp4",
+        "control_weight": 1.0
+    }
+}

output/priority_from_right/run_000/cosmos_input/depth/priority_from_right_right_depth_spec.json

@@ -0,0 +1,10 @@
+{
+    "name": "priority_from_right_right_depth",
+    "prompt_path": "../prompt.txt",
+    "video_path": "../../videos/rgb_right.mp4",
+    "guidance": 3,
+    "depth": {
+        "control_path": "../../videos/depth_right.mp4",
+        "control_weight": 1.0
+    }
+}

output/priority_from_right/run_000/cosmos_input/edge/priority_from_right_front_edge_spec.json

@@ -0,0 +1,10 @@
+{
+    "name": "priority_from_right_front_edge",
+    "prompt_path": "../prompt.txt",
+    "video_path": "../../videos/rgb_front.mp4",
+    "guidance": 3,
+    "edge": {
+        "control_path": "../../videos/edge_front.mp4",
+        "control_weight": 1.0
+    }
+}

output/priority_from_right/run_000/cosmos_input/edge/priority_from_right_left_edge_spec.json

@@ -0,0 +1,10 @@
+{
+    "name": "priority_from_right_left_edge",
+    "prompt_path": "../prompt.txt",
+    "video_path": "../../videos/rgb_left.mp4",
+    "guidance": 3,
+    "edge": {
+        "control_path": "../../videos/edge_left.mp4",
+        "control_weight": 1.0
+    }
+}

output/priority_from_right/run_000/cosmos_input/edge/priority_from_right_rear_edge_spec.json

@@ -0,0 +1,10 @@
+{
+    "name": "priority_from_right_rear_edge",
+    "prompt_path": "../prompt.txt",
+    "video_path": "../../videos/rgb_rear.mp4",
+    "guidance": 3,
+    "edge": {
+        "control_path": "../../videos/edge_rear.mp4",
+        "control_weight": 1.0
+    }
+}

output/priority_from_right/run_000/cosmos_input/edge/priority_from_right_right_edge_spec.json

@@ -0,0 +1,10 @@
+{
+    "name": "priority_from_right_right_edge",
+    "prompt_path": "../prompt.txt",
+    "video_path": "../../videos/rgb_right.mp4",
+    "guidance": 3,
+    "edge": {
+        "control_path": "../../videos/edge_right.mp4",
+        "control_weight": 1.0
+    }
+}

output/priority_from_right/run_000/cosmos_input/multicontrol/priority_from_right_front_multicontrol_spec.json

@@ -0,0 +1,22 @@
+{
+    "name": "priority_from_right_front_multicontrol",
+    "prompt_path": "../prompt.txt",
+    "video_path": "../../videos/rgb_front.mp4",
+    "guidance": 5,
+    "depth": {
+        "control_path": "../../videos/depth_front.mp4",
+        "control_weight": 1.0
+    },
+    "edge": {
+        "control_path": "../../videos/edge_front.mp4",
+        "control_weight": 0.2
+    },
+    "seg": {
+        "control_path": "../../videos/semseg_front.mp4",
+        "control_weight": 1.0
+    },
+    "vis": {
+        "control_weight": 0.3,
+        "preset_blur_strength": "high"
+    }
+}

output/priority_from_right/run_000/cosmos_input/multicontrol/priority_from_right_left_multicontrol_spec.json

@@ -0,0 +1,22 @@
+{
+    "name": "priority_from_right_left_multicontrol",
+    "prompt_path": "../prompt.txt",
+    "video_path": "../../videos/rgb_left.mp4",
+    "guidance": 5,
+    "depth": {
+        "control_path": "../../videos/depth_left.mp4",
+        "control_weight": 1.0
+    },
+    "edge": {
+        "control_path": "../../videos/edge_left.mp4",
+        "control_weight": 0.2
+    },
+    "seg": {
+        "control_path": "../../videos/semseg_left.mp4",
+        "control_weight": 1.0
+    },
+    "vis": {
+        "control_weight": 0.3,
+        "preset_blur_strength": "high"
+    }
+}

output/priority_from_right/run_000/cosmos_input/multicontrol/priority_from_right_rear_multicontrol_spec.json

@@ -0,0 +1,22 @@
+{
+    "name": "priority_from_right_rear_multicontrol",
+    "prompt_path": "../prompt.txt",
+    "video_path": "../../videos/rgb_rear.mp4",
+    "guidance": 5,
+    "depth": {
+        "control_path": "../../videos/depth_rear.mp4",
+        "control_weight": 1.0
+    },
+    "edge": {
+        "control_path": "../../videos/edge_rear.mp4",
+        "control_weight": 0.2
+    },
+    "seg": {
+        "control_path": "../../videos/semseg_rear.mp4",
+        "control_weight": 1.0
+    },
+    "vis": {
+        "control_weight": 0.3,
+        "preset_blur_strength": "high"
+    }
+}

output/priority_from_right/run_000/cosmos_input/multicontrol/priority_from_right_right_multicontrol_spec.json

@@ -0,0 +1,22 @@
+{
+    "name": "priority_from_right_right_multicontrol",
+    "prompt_path": "../prompt.txt",
+    "video_path": "../../videos/rgb_right.mp4",
+    "guidance": 5,
+    "depth": {
+        "control_path": "../../videos/depth_right.mp4",
+        "control_weight": 1.0
+    },
+    "edge": {
+        "control_path": "../../videos/edge_right.mp4",
+        "control_weight": 0.2
+    },
+    "seg": {
+        "control_path": "../../videos/semseg_right.mp4",
+        "control_weight": 1.0
+    },
+    "vis": {
+        "control_weight": 0.3,
+        "preset_blur_strength": "high"
+    }
+}

output/priority_from_right/run_000/cosmos_input/negative_prompt.txt

@@ -0,0 +1 @@
+The video captures a game playing, with bad crappy graphics and cartoonish frames. It represents a recording of old outdated games. The lighting looks very fake. The textures are very raw and basic. The geometries are very primitive. The images are very pixelated and of poor CG quality. There are many subtitles in the footage. Overall, the video is unrealistic at all.

output/priority_from_right/run_000/cosmos_input/prompt.txt

@@ -0,0 +1 @@
+The video shows an urban driving scene captured from a vehicle-mounted camera. Multiple vehicles move through a road network with painted lane markings, stop lines and pedestrian crosswalks. Low-rise buildings, sidewalks, parked cars and street lamps line the streets. Bright daytime lighting with soft shadows; the camera follows the moving ego vehicle smoothly.

output/priority_from_right/run_000/cosmos_input/seg/priority_from_right_front_seg_spec.json

@@ -0,0 +1,10 @@
+{
+    "name": "priority_from_right_front_seg",
+    "prompt_path": "../prompt.txt",
+    "video_path": "../../videos/rgb_front.mp4",
+    "guidance": 3,
+    "seg": {
+        "control_path": "../../videos/semseg_front.mp4",
+        "control_weight": 1.0
+    }
+}

output/priority_from_right/run_000/cosmos_input/seg/priority_from_right_left_seg_spec.json

@@ -0,0 +1,10 @@
+{
+    "name": "priority_from_right_left_seg",
+    "prompt_path": "../prompt.txt",
+    "video_path": "../../videos/rgb_left.mp4",
+    "guidance": 3,
+    "seg": {
+        "control_path": "../../videos/semseg_left.mp4",
+        "control_weight": 1.0
+    }
+}

output/priority_from_right/run_000/cosmos_input/seg/priority_from_right_rear_seg_spec.json

@@ -0,0 +1,10 @@
+{
+    "name": "priority_from_right_rear_seg",
+    "prompt_path": "../prompt.txt",
+    "video_path": "../../videos/rgb_rear.mp4",
+    "guidance": 3,
+    "seg": {
+        "control_path": "../../videos/semseg_rear.mp4",
+        "control_weight": 1.0
+    }
+}

output/priority_from_right/run_000/cosmos_input/seg/priority_from_right_right_seg_spec.json

@@ -0,0 +1,10 @@
+{
+    "name": "priority_from_right_right_seg",
+    "prompt_path": "../prompt.txt",
+    "video_path": "../../videos/rgb_right.mp4",
+    "guidance": 3,
+    "seg": {
+        "control_path": "../../videos/semseg_right.mp4",
+        "control_weight": 1.0
+    }
+}