license: cc-by-nc-4.0
tags:
- image-geolocation
- osv-5m
pipeline_tag: image-feature-extraction
datasets:
- osv5m/osv5m
base_model:
- timm/ViT-SO400M-16-SigLIP2-256
- timm/ViT-gopt-16-SigLIP2-256
- facebook/dinov3-vitl16-pretrain-lvd1689m
model-index:
- name: Chipoint
results:
- task:
type: image-feature-extraction
name: Image Geolocation
dataset:
type: osv5m
name: OSV-5M
metrics:
- name: Acc@25km
type: accuracy
value: 37.63
- name: Acc@200km
type: accuracy
value: 70.56
- name: Acc@750km
type: accuracy
value: 84.94
- name: Acc@2500km
type: accuracy
value: 93.36
- name: GeoScore
type: geoscore
value: 4174
- name: Median error (km)
type: distance-error
value: 51
- name: Mean error (km)
type: distance-error
value: 716
- name: Country accuracy
type: accuracy
value: 85.55
- name: Region accuracy
type: accuracy
value: 61.68
- task:
type: image-feature-extraction
name: Image Geolocation (cross-dataset generalization)
dataset:
type: geoguessr
name: GeoGuessr (Google Street View)
metrics:
- name: Country accuracy
type: accuracy
value: 63.14
Chipoint is a SOTA geolocation model that is open-weight and currently beats the published frontier closed-weight model (Pinpoint, June 2026) and beats all models on the OSV-5M bench.
Chipoint — OSV-5M benchmark
OSV-5M test set (210,122 images). Acc@Xkm is the share of predictions within X km. GeoScore is 5000·exp(-d/1492.7), higher is better.
| Model | @25km | @200km | @750km | @2500km | GeoScore | Mean err |
|---|---|---|---|---|---|---|
| OSV-5M Baseline | – | – | – | – | 3361 | 1814 km |
| GeoCLIP | 21.5 | 52.1 | 72.1 | – | – | – |
| GRE | 9.7 | 35.6 | 72.5 | 91.1 | – | 1192 km |
| LocDiff | 11.0 | 46.3 | 77.0 | 88.2 | – | – |
| RFM (S2S) | – | – | – | – | 3767 | 1069 km |
| HierLoc | – | – | – | – | 3963 | 861 km |
| Pinpoint (retrieval only) | 32.1 | 65.6 | 82.8 | 92.8 | 4035 | 784 km |
| Pinpoint (full) | 35.6 | 67.5 | 83.7 | 93.2 | 4114 | 743 km |
| Chipoint | 37.63 | 70.56 | 84.94 | 93.36 | 4174 | 716 km |
Chipoint's median error is 51 km. Baseline numbers are as reported in the OSV-5M (CVPR 2024), HierLoc (arXiv 2601.23064), and Pinpoint (arXiv 2606.04133) papers. Blank cells are metrics a given method does not report on OSV-5M. OSV-5M Baseline, RFM, and HierLoc report administrative accuracy plus GeoScore/mean rather than distance thresholds, so they compare only on the GeoScore and mean-error columns.
Country / region (OSV-5M): Chipoint reaches country ≈ 85.6%, region / admin-1 ≈ 61.7% (from reverse-geocoding the predictions, approximate vs exact admin boundaries). For reference, at admin level Pinpoint reports Country 84.8% / Region 59.1% and HierLoc 82.9% / 55.0%. Our figures use nearest-place geocoding rather than exact polygons, so treat that comparison as indicative.
IMPORTANT
Chipoint could not compare itself on other benchmarks other than OSV-5M due to several factors. Benchmarks such as im2gps, im2gps3k, YFCC4k, and GWS15k are either private or unretrivable.
Running the model
Instructions on how to run Chipoint locally are located in the RUN_GUIDE.md file.
when the hell is chipoint 2 coming out??
its in progress! we are cooking at chiikabu labs, and if youd like to accelerate chipoint 2's progress, feel free to contact me on discord! (username is "fourlatte")