“YORU” is an open-source sftware for animal behavior detection using an object detection. YORU classifies animal behavior as "behavior objects."
Support the development of YORU.
Sponsor©Copyright 2026 Hayato M. Yamanouchi
When you report a problem on GitHub Issues, please attach the log file
~/.yoru/logs/yoru.log(%USERPROFILE%\.yoru\logs\yoru.logon Windows). From v1.1.2 and v2.0.0-beta.3 it holds the full traceback of every error that the GUIs catch.
Q. The launcher stops with EnvironmentError: Can't find Google Chrome/Chromium installation. (v1.1.x)
A. The v1.1.x launcher opens in Google Chrome or Chromium. Install one of them. Microsoft Edge and Safari are not detected. The v2.0 betas do not need a browser.
Q. Do I need to install the CUDA toolkit?
A. No. Only the NVIDIA driver is needed (527.41 or newer on Windows, 525.60.13 or newer on Linux). The PyTorch wheels carry their own CUDA runtime. Check the driver with nvidia-smi. The CUDA 12.8 build that v2.0.0-beta.4 uses needs driver 570 or newer.
Q. A YORU screen closes immediately after it opens, and nothing is written to yoru.log.
A. An error while a screen is starting up can close its console before it is logged. Run that screen directly from a terminal to see the error, for example python -m yoru.train_GUI, and always launch YORU from the repository root.
Q. import cv2 fails after I installed or upgraded packages myself.
A. opencv-python 4.10 is built against NumPy 1.x, so NumPy 2 breaks it. YORU’s environment files cap NumPy below 2. If NumPy 2 was pulled in by another package, run pip install "numpy<2" in the YORU environment (or run uv sync again with uv).
Q. uv sync fails on macOS while compiling a package.
A. Some dependencies have no Apple Silicon wheels and are compiled during the sync. Install the Xcode Command Line Tools first with xcode-select --install, then run uv sync again.
Q. Can I keep using my Beta 2 conda environment for v2.0.0-beta.3?
A. No. Beta 3 moved from Python 3.9 to 3.10, and updating an environment in place is not reliable across a Python version change. Recreate the environment, as described in the Beta Install guide. uv users only need to run uv sync.
Q. Can I keep using my Beta 3 conda environment for v2.0.0-beta.4?
A. Yes. Python stays at 3.10 and YORU.yml has not changed. Install the CUDA 12.8 build of PyTorch in it, as described in the Beta Install guide. uv users only need to run uv sync.
Q. CUDA calls fail with no kernel image is available on an RTX 50-series (Blackwell) GPU.
A. The PyTorch build has no kernels for these cards. v2.0.0-beta.4 uses the CUDA 12.8 build on both routes, which supports them: with uv, check out the tag and run uv sync; with conda, install it as in the Beta Install guide. It needs driver 570 or newer. With v1.1.2 or an earlier beta on uv, whose lockfile pins a CUDA 12.4 build, set YORU_DEVICE=cpu, or use the conda route and install a CUDA 12.8 build of PyTorch.
Q. How do I choose between GPU and CPU?
A. From v1.1.2, YORU picks CUDA, then Apple MPS, then the CPU. To choose yourself, set YORU_DEVICE (cuda, mps, cpu, or a CUDA index such as 0) before launching, or use the Device selector in the Training GUI. See Install.
Q. Real-time detection on my Mac is slower on MPS than I expected.
A. MPS speeds up training, but for real-time inference with the small YOLO models it is not faster than the CPU. Try YORU_DEVICE=cpu.
Q. We found “torch.cuda.OutOfMemoryError: CUDA out of memory” on the terminal.
A. This occurs when the mini-batch during training exceeds the GPU’s memory capacity. Try a smaller batch size. In the v2.0 betas, the Training GUI estimates GPU memory before a run starts and offers the largest batch expected to fit.
Q. LabelImg from v1.1.x no longer starts after I used the v2.0 beta.
A. Both versions share the settings file ~/.labelImgSettings.pkl. After the beta’s bundled LabelImg has saved its settings there, the LabelImg used by v1.1.x can fail to start. Delete or rename ~/.labelImgSettings.pkl (%USERPROFILE%\.labelImgSettings.pkl on Windows). It only stores LabelImg’s preferences, not your labels.
Q. OBB training stops with OBB dataset incorrectly formatted. (v2.0.0-beta.3 and later)
A. The dataset mixes ordinary (5-field) and OBB (9-field) label files. In an OBB session, opening an ordinary label file switches LabelImg to plain YOLO, and boxes saved after that lose their angle. Relabel those files as rotated boxes, or remove them from the dataset. See Beta: Training.
Q. YORU fails to load YOLO weights with a weights_only / Weights only load failed error.
A. The conda YORU.yml of v1.1.x pins ultralytics==8.2.52, which can fail with PyTorch 2.6+ because the default for torch.load changed to weights_only=True (PyTorch documentation).
python -m pip install -U ultralytics and restart YORU.uv sync, then uv run yoru from its folder. Its pyproject.toml requires ultralytics>=8.3.0, excluding 8.2.52; uv sync uses the project’s lockfile.The cause is the dependency versions, not the choice of pip, conda, or uv.
Q. My YOLOv5 model does not load in the v2.0 beta.
A. Use v2.0.0-beta.4 or later. Betas 2 and 3 could not load YOLOv5 .pt weights; Beta 4 bundles YOLOv5 again, loads models trained with YORU v1, and gives the same boxes as v1. Leave yolo_model_type on auto, or set it to yolov5. See Beta: Training.
Q. The TTL signal does not come out of the pin set in trigger_pin. (v1.1.x)
A. v1.1.x always outputs the TTL signal on Arduino digital pin 13 and does not apply trigger_pin or trigger_threshold_configuration. Wire the output to pin 13. Both settings are applied in the v2.0 betas.