Downloads · 30 days
0
josefbednar/neuroguessr-app
neuroguessr-app is a machine learning model from josefbednar. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as other.
Point it at a photograph and it tells you where on Earth it was taken. Everything runs on the phone — the neural encoder, the 300,000-place index, and the map. After the one-time asset download below, the app never to…
Downloads · 30 days
0
Access
Public
Updated Aug 1, 2026
Repo size
2.3 GB
Likes
0
Public
Click a slice to open those files.
.i81.5 GB · 68%
From the Hugging Face model README
Point it at a photograph and it tells you where on Earth it was taken. Everything runs on the phone — the neural encoder, the 300,000-place index, and the map. After the one-time asset download below, the app never touches the network.
<img src="screenshot.png" width="320" alt="NeuroGuessr showing a prediction in Serbia">You need an arm64 phone with Android 10 or newer and about 4.5 GB free.
If an older build is already installed and the update fails with "App not installed", uninstall the old one first — early builds carried a different signature.
neuroguessr-0.2.apk — the app itself (Kotlin + Compose + ONNX Runtime).assets/model/encoder_fp16.onnx — a DINOv3 ViT-L/16 backbone, contrastively fine-tuned for
place recognition, exported at fp16 with the residual stream kept fp32.assets/index/ — 299,518 places as int8-quantised descriptors in five learned spaces,
cell-sorted for a gated search.assets/download_manifest.json — sizes and SHA-256 of every asset; the copy baked into the
APK is what the in-app downloader verifies against.Median error on the held-out validation set: 44.4 km, with 43% of guesses within 25 km.
The encoder is a derivative of facebook/dinov3-vitl16-pretrain-lvd1689m and is distributed
under the DINOv3 licence, a copy of which ships in this repo. Third-party
notices are in NOTICE.md. Source and asset pipeline:
the app is built from the NeuroGuessr mobile repo; assets regenerate from the research repo.