[Bug]:

[Tommaswang]

Bug Description

run python script ：
1、python scripts/eval/start_server.py --config scripts/eval/configs/h1_internvla_n1_cfg.py

docker：
2、 MESA_GL_VERSION_OVERRIDE=4.6 python scripts/eval/eval.py --config scripts/eval/configs/h1_internvla_n1_cfg.py

error：
2025-12-06 08:29:33 [14,334ms] [Error] [omni.kit.app._impl] [py stderr]: Traceback (most recent call last):
File "/root/InternNav/scripts/eval/eval.py", line 44, in
main()
File "/root/InternNav/scripts/eval/eval.py", line 40, in main
evaluator.eval()
File "/root/InternNav/internnav/evaluator/vln_multi_evaluator.py", line 308, in eval
self.visualize_util.save_observation(
File "/root/InternNav/internnav/evaluator/utils/visualize_util.py", line 132, in save_observation
self._save_frame_fn(ti.saved_frames, action, out_path, ti.reference_path)
File "/root/InternNav/internnav/evaluator/utils/common.py", line 644, in obs_to_image
array = np.concatenate((rgb_array, topdown_array), axis=1)
ValueError: all the input array dimensions except for the concatenation axis must match exactly, but along dimension 0, the array at index 0 has size 480 and the array at index 1 has size 256

anyone has encounter problems like this ？？？

Steps to Reproduce

Please try to provide a minimal example to reproduce the bug. Error messages and stack traces are also helpful.

Expected Behavior

run csript without error！！！

Screenshots/Videos

No response

Environment

• OS: [e.g. Ubuntu 22.04]
• GPU/CPU [e.g. A100, RTX 4090, i9-14900K]
• GPU-driver version

Release version or Commit ID

Name: internnav
Version: 0.0.1
Summary: InternNav: A benchmark evaluation framework for navigation tasks
Home-page:
Author: Intern Robotics
Author-email: embodiedai@pjlab.org.cn
License: Apache 2.0
Location: /home/ubuntu/miniconda3/envs/internnav/lib/python3.10/site-packages
Requires: ansi2txt, anyio, coloredlogs, coverage, fastapi, filelock, fsspec, gpustat, gunicorn, gym, gym-notices, gymnasium, httpx, imageio, imageio-ffmpeg, ipdb, jsonschema, jsonschema-specifications, msgpack-numpy, numpy, nvsmi, opencv-python-headless, packaging, pillow, portalocker, propcache, psutil, pydantic, pytest, pytest-cov, pytest-timeout, python-dateutil, python-multipart, PyYAML, ray, referencing, requests, rich, rpds-py, sentry-sdk, setproctitle, sniffio, starlette, termcolor, tomli, tqdm, typing_extensions, tyro, uvicorn, watchdog
Required-by:

Additional Context

No response

[LeslieLaw0925]

The following two ways can temporarily solve the problem.
1、Change the camera_resolution of TaskCfg to [256, 256] in scripts/eval/configs/h1_internvla_n1_cfg.py.
2、Or, you can add rgb_array = crop(rgb_array) at line 614 in internnav/evaluator/utils/common.py.

But It's unclear whether this change will affect the code logic.

[kew6688]

Hi, we’ve already fixed this bug in #168.
The issue was caused by a resolution mismatch between the camera and the top-down camera, which is only used for visualization purposes.

Thanks @LeslieLaw0925 for providing the solution — it’s exactly the right fix to avoid this dimension mismatch problem.

[Tommaswang]

can you fix it in config files or change the code to support both [640,480] 、[256,256] ??

[kew6688]

Yes, the latest code supports both [640,480] 、[256,256]
this bug is fixed in PR #168.

Updated:
You can fix it by update this line in internnav/evaluator/utils/common.py

[Tommaswang]

OK ！ problem soloved ！！