Tune quickstart capture framing and downsampling for documentation media

The quickstart GIFs rendered the OVRTX path tracer straight to their 320x240
publication size, so robot silhouettes and shadow edges aliased badly, and the
per-task focal lengths left the subject occupying a small part of the frame
with a large empty floor area around it.

Capture at 1280x960 and let the generator downsample with lanczos, which
resolves the detail that rendering directly to the final size destroyed. Tighten
the focal lengths per task to the largest value that keeps the subject inside
the published crop. The safe value is scene-specific rather than a fixed ratio:
Cartpole clips its pole above 21, and Kuka Allegro is held at 24 because its
lift policy raises the arm above the pose a zero action holds.

Ease the denoiser from 2:2:6:6 to 1:1:3:3 now that supersampling rather than
blurring removes the path-tracer noise, and widen the GIF palette from 128 to
256 colors with sierra2_4a dithering to avoid banding on the smoother gradients.

Author: mataylor-nvidia

tools/docs/media/capture_quickstart.py

@@ -29,6 +29,8 @@
 from isaaclab_tasks.core.velocity.config.g1.flat_env_cfg import G1FlatEnvCfg
 from isaaclab_tasks.utils import resolve_task_config, setup_preset_cli
 
+# Focal lengths are tuned per task to the tightest framing that still keeps the subject inside the
+# published crop; the safe value differs per scene, so they are not a fixed ratio of one another.
 _TASK_CONFIGS = {
     "Isaac-Cartpole": "capture_quickstart:CartpoleCaptureCfg",
     "Isaac-Lift-KukaAllegro": "capture_quickstart:KukaAllegroCaptureCfg",
@@ -43,7 +45,7 @@ class CartpoleCaptureCfg(CartpoleEnvCfg):
 
     def __post_init__(self):
         super().__post_init__()
-        _configure_capture(self, focal_length=18.0)
+        _configure_capture(self, focal_length=21.0)
 
 
 @configclass
@@ -52,7 +54,7 @@ class G1FlatCaptureCfg(G1FlatEnvCfg):
 
     def __post_init__(self):
         super().__post_init__()
-        _configure_capture(self, focal_length=20.0)
+        _configure_capture(self, focal_length=29.0)
 
 
 @configclass
@@ -61,7 +63,9 @@ class KukaAllegroCaptureCfg(KukaAllegroLiftEnvCfg):
 
     def __post_init__(self):
         super().__post_init__()
-        _configure_capture(self, focal_length=22.0)
+        # Held to 24 rather than the tighter framing the other tasks take: the lift policy raises
+        # the arm well above the pose a zero action holds, and 26 already leaves little headroom.
+        _configure_capture(self, focal_length=24.0)
 
 
 @configclass
@@ -70,7 +74,7 @@ class FrankaCabinetCaptureCfg(FrankaCabinetEnvCfg):
 
     def __post_init__(self):
         super().__post_init__()
-        _configure_capture(self, focal_length=22.0)
+        _configure_capture(self, focal_length=32.0)
 
 
 def configure_playback() -> list[str]:
@@ -97,7 +101,7 @@ def main():
     env_cfg, _ = resolve_task_config(args.task, "", play_mode=True)
     env_cfg.scene.num_envs = 1
     env_cfg.seed = 42
-    _configure_resolved_sim(env_cfg.sim, focal_length=22.0)
+    _configure_resolved_sim(env_cfg.sim, focal_length=32.0)
     env_cfg.video_recorders = [
         VideoRecorderCfg(
             source="visualizer:newton_rtx",
@@ -144,15 +148,20 @@ def _configure_capture(env_cfg: object, focal_length: float):
 
 
 def _configure_resolved_sim(sim_cfg: SimulationCfg, focal_length: float):
-    """Attach a headless 320 by 240 OVRTX visualizer to a simulation config."""
+    """Attach a headless 1280 by 960 OVRTX visualizer to a simulation config.
+
+    Captures at four times the 320 by 240 publication size so the generator can downsample the
+    path-traced frames, which resolves robot silhouettes and shadow edges that alias away when
+    the path tracer renders straight to the final size.
+    """
     default_camera = sim_cfg.default_visualizer_cfg or VisualizerCfg()
     sim_cfg.visualizer_cfgs = [
         NewtonRTXVisualizerCfg(
             eye=default_camera.eye,
             lookat=default_camera.lookat,
             focal_length=focal_length,
-            window_width=320,
-            window_height=240,
+            window_width=1280,
+            window_height=960,
             headless=True,
             rtx_environment="default",
             render_settings={"omni:rtx:quality": ("Int", 100)},

tools/docs/media/generate_quickstart.sh

@@ -75,25 +75,25 @@ ffmpeg -y -v error \
     -i "${WORK_DIR}/random/random_0000.mp4" \
     -i "${WORK_DIR}/play/play_0000.mp4" \
     -filter_complex \
-    "[0:v]select='not(mod(n,5))',setpts=N/(12*TB),hqdn3d=2:2:6:6,drawbox=x=8:y=8:w=112:h=28:color=black@0.65:t=fill,drawtext=font='DejaVu Sans':text='zero_agent':fontcolor=white:fontsize=16:x=16:y=13[zero];
-     [1:v]select='not(mod(n,5))',setpts=N/(12*TB),hqdn3d=2:2:6:6,drawbox=x=8:y=8:w=134:h=28:color=black@0.65:t=fill,drawtext=font='DejaVu Sans':text='random_agent':fontcolor=white:fontsize=16:x=16:y=13[random];
-     [2:v]select='not(mod(n,5))',setpts=N/(12*TB),hqdn3d=2:2:6:6,drawbox=x=8:y=8:w=58:h=28:color=black@0.65:t=fill,drawtext=font='DejaVu Sans':text='play':fontcolor=white:fontsize=16:x=16:y=13[play];
+    "[0:v]select='not(mod(n,5))',setpts=N/(12*TB),scale=320:240:flags=lanczos,hqdn3d=1:1:3:3,drawbox=x=8:y=8:w=112:h=28:color=black@0.65:t=fill,drawtext=font='DejaVu Sans':text='zero_agent':fontcolor=white:fontsize=16:x=16:y=13[zero];
+     [1:v]select='not(mod(n,5))',setpts=N/(12*TB),scale=320:240:flags=lanczos,hqdn3d=1:1:3:3,drawbox=x=8:y=8:w=134:h=28:color=black@0.65:t=fill,drawtext=font='DejaVu Sans':text='random_agent':fontcolor=white:fontsize=16:x=16:y=13[random];
+     [2:v]select='not(mod(n,5))',setpts=N/(12*TB),scale=320:240:flags=lanczos,hqdn3d=1:1:3:3,drawbox=x=8:y=8:w=58:h=28:color=black@0.65:t=fill,drawtext=font='DejaVu Sans':text='play':fontcolor=white:fontsize=16:x=16:y=13[play];
      [zero][random][play]hstack=inputs=3,split[frames][palette_frames];
-     [palette_frames]palettegen=max_colors=128:stats_mode=diff[palette];
-     [frames][palette]paletteuse=dither=bayer:bayer_scale=4:diff_mode=rectangle" \
+     [palette_frames]palettegen=max_colors=256:stats_mode=diff[palette];
+     [frames][palette]paletteuse=dither=sierra2_4a:diff_mode=rectangle" \
     -loop 0 "${OUTPUT_DIR}/agent-comparison.gif"
 
 ffmpeg -y -v error \
     -i "${WORK_DIR}/cartpole/cartpole_0000.mp4" \
     -i "${WORK_DIR}/g1/g1_0000.mp4" \
     -i "${WORK_DIR}/kuka_allegro/kuka_allegro_0000.mp4" \
     -filter_complex \
-    "[0:v]crop=320:180:0:30,select='not(mod(n,5))',setpts=N/(12*TB),hqdn3d=2:2:6:6,drawbox=x=8:y=8:w=82:h=28:color=black@0.65:t=fill,drawtext=font='DejaVu Sans':text='Cartpole':fontcolor=white:fontsize=16:x=16:y=13[cartpole];
-     [1:v]crop=320:180:0:30,select='not(mod(n,5))',setpts=N/(12*TB),hqdn3d=2:2:6:6,drawbox=x=8:y=8:w=128:h=28:color=black@0.65:t=fill,drawtext=font='DejaVu Sans':text='G1 locomotion':fontcolor=white:fontsize=16:x=16:y=13[g1];
-     [2:v]crop=320:180:0:30,select='not(mod(n,5))',setpts=N/(12*TB),hqdn3d=2:2:6:6,drawbox=x=8:y=8:w=116:h=28:color=black@0.65:t=fill,drawtext=font='DejaVu Sans':text='Kuka Allegro':fontcolor=white:fontsize=16:x=16:y=13[kuka];
+    "[0:v]crop=1280:720:0:120,select='not(mod(n,5))',setpts=N/(12*TB),scale=320:180:flags=lanczos,hqdn3d=1:1:3:3,drawbox=x=8:y=8:w=82:h=28:color=black@0.65:t=fill,drawtext=font='DejaVu Sans':text='Cartpole':fontcolor=white:fontsize=16:x=16:y=13[cartpole];
+     [1:v]crop=1280:720:0:120,select='not(mod(n,5))',setpts=N/(12*TB),scale=320:180:flags=lanczos,hqdn3d=1:1:3:3,drawbox=x=8:y=8:w=128:h=28:color=black@0.65:t=fill,drawtext=font='DejaVu Sans':text='G1 locomotion':fontcolor=white:fontsize=16:x=16:y=13[g1];
+     [2:v]crop=1280:720:0:120,select='not(mod(n,5))',setpts=N/(12*TB),scale=320:180:flags=lanczos,hqdn3d=1:1:3:3,drawbox=x=8:y=8:w=116:h=28:color=black@0.65:t=fill,drawtext=font='DejaVu Sans':text='Kuka Allegro':fontcolor=white:fontsize=16:x=16:y=13[kuka];
      [cartpole][g1][kuka]hstack=inputs=3,split[frames][palette_frames];
-     [palette_frames]palettegen=max_colors=128:stats_mode=diff[palette];
-     [frames][palette]paletteuse=dither=bayer:bayer_scale=4:diff_mode=rectangle" \
+     [palette_frames]palettegen=max_colors=256:stats_mode=diff[palette];
+     [frames][palette]paletteuse=dither=sierra2_4a:diff_mode=rectangle" \
     -loop 0 "${OUTPUT_DIR}/task-sampler.gif"
 
 for output in agent-comparison.gif task-sampler.gif; do