Quickstart Guide Fixes

README.md

@@ -160,14 +160,17 @@ ns-train nerfacto --vis viewer --viewer.zmq-port 8001 --viewer.websocket-port 80
 
 ### 3.x Training a model with the viewer
 
-Make sure to forward a port for the websocket to localhost. The default port is 7007, which you should be expose to localhost:7007.
+Make sure to forward a port for the websocket to localhost. The default port is 7007, which you should expose to localhost:7007.
 
 ```bash
 # with the default port
 ns-train nerfacto --vis viewer
 
 # with a specified websocket port
 ns-train nerfacto --vis viewer --viewer.websocket-port=7008
+
+# port forward if running on remote
+ssh -L localhost:7008:localhost:7008 {REMOTE HOST}
 ```
 
 ### 4. Visualizing training runs
@@ -201,10 +204,10 @@ We support logging to weights and biases. To enable wandb logging, add the flag
 
 </details>
 
-## 5. Rendering a trajectories during inference
+## 5. Rendering a trajectory during inference
 
 ```
-ns-eval render-trajectory --load-config=outputs/blender_lego/instant_ngp/2022-07-07_230905/config.yml--traj=spiral --output-path=output.mp4
+ns-eval render-trajectory --load-config=outputs/blender_lego/instant_ngp/{MOST_RECENT_RUN}/config.yml--traj=spiral --output-path=output.mp4
 ```
 
 ## 6. In-depth guide

nerfstudio/models/vanilla_nerf.py

@@ -194,8 +194,6 @@ def get_image_metrics_and_images(
             "fine_psnr": float(fine_psnr),
             "fine_ssim": float(fine_ssim),
             "fine_lpips": float(fine_lpips),
-            "ray_loss_coarse": float(torch.mean(outputs["ray_loss_coarse"])),
-            "ray_loss_fine": float(torch.mean(outputs["ray_loss_fine"])),
         }
         images_dict = {"img": combined_rgb, "accumulation": combined_acc, "depth": combined_depth}
         return metrics_dict, images_dict

scripts/eval.py

@@ -8,6 +8,7 @@
 import json
 import logging
 import os
+import sys
 from dataclasses import dataclass
 from pathlib import Path
 from typing import Literal, Tuple, Union
@@ -48,6 +49,14 @@ def _load_checkpoint(config: cfg.TrainerConfig, pipeline: Pipeline) -> Path:
     if config.load_step is None:
         console.print("Loading latest checkpoint from load_dir")
         # NOTE: this is specific to the checkpoint name format
+        if not os.path.exists(config.load_dir):
+            console.rule("Error", style="red")
+            console.print(f"No checkpoint directory found at {config.load_dir}, ", justify="center")
+            console.print(
+                "Please make sure the checkpoint exists, they should be generated periodically during training",
+                justify="center",
+            )
+            sys.exit(1)
         load_step = sorted(int(x[x.find("-") + 1 : x.find(".")]) for x in os.listdir(config.load_dir))[-1]
     else:
         load_step = config.load_step
@@ -93,6 +102,11 @@ def _render_trajectory_video(
             camera_ray_bundle = cameras.generate_rays(camera_indices=camera_idx).to(pipeline.device)
             with torch.no_grad():
                 outputs = pipeline.model.get_outputs_for_camera_ray_bundle(camera_ray_bundle)
+            if rendered_output_name not in outputs:
+                console.rule("Error", style="red")
+                console.print(f"Could not find {rendered_output_name} in the model outputs", justify="center")
+                console.print(f"Please set --rendered_output_name to one of: {outputs.keys()}", justify="center")
+                sys.exit(1)
             image = outputs[rendered_output_name].cpu().numpy()
             images.append(image)