Display network-forward throughput FPS and allow up to 300 frames

README.md

@@ -20,6 +20,7 @@ The **Filter sky** control is enabled by default and available under Advanced.
 Reconstruction displays and logs **Network forward pass** time with CUDA
 synchronized before and after the network call. Loading, preprocessing,
 prediction postprocessing, sky filtering and file export are excluded.
+Displayed throughput FPS is input frame count divided by network forward seconds.
 Visualization-only changes retain the last timing; cached examples do not
 run the network or display a newly measured time.
 
@@ -30,7 +31,7 @@ private. Visibility is a Space setting, not controlled by this README.
 ## Uploads and temporary results
 
 Live inference accepts JPEG, PNG or WebP images and supported video files,
-up to 128 images or sampled video frames, at most
+up to 300 images or sampled video frames, at most
 16 megapixels per frame. Uploads are limited to 100 MB per file at the
 HTTP upload boundary and 256 MiB total per submitted collection. Videos
 are limited to 900 decoded frames and sampled at 0.5–10 FPS.

src/dvlt/scripts/gradio_demo.py

@@ -79,7 +79,7 @@ from dvlt.viz.scene_rerun import visualize_scene
 
 
 logger = get_logger(__name__)
-HOSTED_MAX_FRAMES = 128
+HOSTED_MAX_FRAMES = 300
 
 
 EXTRA_CLI_ARGS = [
@@ -418,14 +418,15 @@ def build_app(
         return path
 
     def _invalidate_inference(state):
-        for key in ("predictions", "batch", "pts", "rgb", "forward_seconds", "inference_model"):
+        for key in ("predictions", "batch", "pts", "rgb", "forward_seconds", "forward_frame_count", "inference_model"):
             state.pop(key, None)
 
     def _timing_notice(state):
         seconds = state.get("forward_seconds")
         if seconds is None:
             return ""
-        return f"**Network forward pass: {seconds:.3f} s** · {state['inference_model']}"
+        fps = f"{state['forward_frame_count'] / seconds:.2f} FPS" if seconds > 0 else "FPS unavailable"
+        return f"**Network forward pass: {seconds:.3f} s · {fps}** · {state['inference_model']}"
 
     def _model_status(label):
         entry = registry[label]
@@ -541,6 +542,7 @@ def build_app(
             with torch.no_grad(), accelerator.autocast():
                 predictions, forward_seconds = entry.predict_timed(batch, accelerator)
         state["forward_seconds"] = forward_seconds
+        state["forward_frame_count"] = len(frames)
         state["inference_model"] = entry.label
         if os.getenv("SPACES_ZERO_GPU"):
             from dvlt.scripts.cache_full_examples import cpu_predictions