{"id":"abf23fb8-e442-4c31-bd8c-245534729e9e","companyId":"02fb7381-464b-430e-91c5-6a11cf8d2572","modelName":"Carver21","slug":"deepscale-carver21","description":"Carver21 was DeepScale's modular deep learning perception software for driver-assistance and autonomous vehicles. It used three parallel neural networks for object detection, lane identification, and drivable area identification, running at 0.6 tera-ops/sec on low-cost automotive-grade processors.","formFactor":"brain","industryVertical":null,"environment":null,"maturityStage":"commercial","lifecycleState":"active","supersededByModelId":null,"specs":{},"manufacturerSerial":null,"reviewStatus":"reviewed","sources":[{"url":"https://en.wikipedia.org/wiki/DeepScale","label":"DeepScale - Wikipedia","title":"DeepScale - Wikipedia","mediaType":"article","sourceName":"Wikipedia"},{"url":"https://www.deeplearning.ai/the-batch/issue-8","date":"2019-10-09","title":"Tesla Bets on Slim Neural Nets","sourceName":"The Batch / DeepLearning.AI"}],"aliases":[],"collisionRisk":"low","reviewNote":null,"manufacturerTermForTeleop":null,"keyFacts":[{"label":"Perception stack (reported, Oct 2019)","value":"Uses SqueezeNet and three parallel networks for object detection, lane identification and drivable-area identification, per The Batch."},{"label":"Compute budget (reported, Oct 2019)","value":"Runs within 0.6 trillion operations per second, a small share of the 36 trillion attributed to Tesla's custom chipset."},{"label":"DeepScale background (reported, Oct 2019)","value":"Founded in 2015 by two UC Berkeley computer scientists; raised nearly $19 million before Tesla acquired it."}],"createdAt":"2026-08-03T23:04:34.518Z","updatedAt":"2026-10-10T20:40:48.962Z","unitsInFieldClaimed":null,"unitsInFieldVerified":null,"unitsManufacturedClaimed":null,"unitsManufacturedVerified":null,"unitsAsOf":null,"unitsSource":null,"salesModel":null,"priceUndisclosedSince":null,"autonomyClass":null,"autonomyStatement":null,"autonomyStatementSource":null,"demoForensics":null,"exportControlClass":null,"minCommitment":null,"warrantyTerms":null,"serviceModel":null,"productionRateClaimed":null,"productionRateEvidenced":null,"teleopRatioDisclosed":null,"interventionRateDisclosed":null,"jsonLd":{"@context":"https://schema.org","@type":"Product","@id":"https://registry.deploy.report/models/deepscale-carver21","url":"https://registry.deploy.report/models/deepscale-carver21","name":"Carver21","description":"Carver21 was DeepScale's modular deep learning perception software for driver-assistance and autonomous vehicles. It used three parallel neural networks for object detection, lane identification, and drivable area identification, running at 0.6 tera-ops/sec on low-cost automotive-grade processors.","identifier":"abf23fb8-e442-4c31-bd8c-245534729e9e","category":"brain","publisher":{"@id":"https://deploy.report/#organization"}},"framework_metadata":{"framework_schema_version":"0.1.0","verification_status":"verified","maturity_stage":"commercial","lifecycle_state":"active","architectural_position":{"cohort":"brain","sub_cohorts":[]},"within_cohort_verified_vs_claimed_pair":null,"cap_flags":[],"verification_depth":{"sources_count":2,"primary_source_types":["knowledge-base"]}}}