Shofo

Access proprietary multimodal datasets to train and evaluate machine learning models

Pricing model

Demo

Time to value

Moderate Setup (1-3 hours)

You can use Shofo to source large volumes of multimodal training data for machine learning models. The catalog includes millions of hours of proprietary recordings across computer usage, talking head videos, longform speech, video gameplay, and skilled trade labor. For computer use and gaming, datasets include screen recordings paired with synchronized mouse and keyboard inputs. You can also access conversational audio with transcripts or close-up recordings of surgical and carpentry work, allowing research teams to evaluate, pretrain, and fine-tune multimodal systems effectively.

What Shofo does

Request multimodal training data samples for AI evaluationInspect screen recordings synchronized with mouse clicks and keystrokesFilter talking head video clips by specific facial occlusionsDownload longform audio paired with accurate verbatim transcriptsAnalyze physical trade tasks through close-up surgical and carpentry videosAcquire expert gameplay logs to train reinforcement learning agentsFull input logs capturing mouse and keyboard actions paired with screen videoExpert computer use data across CAD, Photoshop, and research tasksFilterable talking head video collections tagged by face occlusions and objectsLongform conversational podcasts and interviews paired with complete transcriptsSynchronized video game recordings with gameplay input captureClose-up recordings of expert trades like surgery and carpentrySelf-serve interface to browse and inspect multimodal data samples