LogitAtlasStreet-scene perception benchmark
Kyoto Vision Labs: infrastructure and git (1.5 years)
Data mix
approximate, based on what's included- Code
8.4K framesVolume
1.5 years (2024–2026)Time span
Computer Vision · 11–50 employeesCompany profile
Ready nowAvailability
Environment in Computer Vision
Kyoto Vision Labs · 11–50 employees · 1.5 years · 8.4K frames
Daily pick
One of today's picks from our strongest environments. Picks refresh every day.
K
Listed by Kyoto Vision Labs
Verified through Stripe Connect. Based in Japan.
Full inventory ready
The owner has shared a file-level inventory, so we can send you the schema and samples quickly.
De-identified before delivery
Reviewer-tested release
Owner-approved licence
The owner signs off on each buyer, scope and term in writing.
Held-out night and rain frames with instance masks for autonomous perception models.
A benchmark split with 8,400 frames never released in the public corpus, labelled by the same annotation team. Includes an evaluation harness that scores segmentation, detection, and tracking with per-condition breakdowns.
What this environment offers
- 8,400 frames captured over 2 years
- Docker export with full history and metadata
- AWS export with full history and metadata
- GitHub export with full history and metadata
- Consistent identifiers so sources can be joined
Stack
- Docker
- AWS
- GitHub
Good for
Agent evaluationReinforcement learningEnterprise / business agentsClassification
The details
SectorComputer Vision
Team size11–50 employees
OriginBuilt from first-party records
FormatsContainer images, Git repositories
DeliveryFile transfer
LanguagesJapanese, English
Label typesFolder structure and filed decisions
Personal dataNone
De-identificationReviewer-tested release
RightsOwner-held
Still growingNo
SampleOn request
RegionEast Asia · Japan
ReferenceDM-GTIRYP
Tags
benchmarksegmentationheld-out