Baidu de­vel­ops open source deep learn­ing frame­work for smart­phones

OpenSource For You - - Fossbytes -

Chi­nese In­ter­net gi­ant Baidu aims to get big­ger in the world of ar­ti­fi­cial in­tel­li­gence (AI) by launch­ing its open source mo­bile deep learn­ing frame­work. Called Mo­bile

Deep Learn­ing (MDL), this con­vo­lu­tion-based neu­ral net­work uses the im­age sen­sor on An­droid and iOS mo­bile de­vices to recog­nise ob­jects.

MDL’s neu­ral net­work has low com­plex­ity and high speed. In­ter­est­ingly, the new model is also de­signed to use the GPU on iOS de­vices to en­able cal­cu­la­tions through the net­work. This takes Ap­ple de­vices a step ahead of the An­droid mod­els.

A cou­ple of demon­stra­tions put up by Baidu on the GitHub repos­i­tory show its dif­fer­ent fea­tures. The first demon­stra­tion re­veals that MDL is ca­pa­ble of recog­nis­ing an ob­ject in less than a sec­ond. The sec­ond demo shows the project’s abil­ity to iden­tify an ob­ject from an im­age and find suit­able matches from the Web.

While MDL does sup­port devel­op­ments made on UC Berke­ley’s deep learn­ing frame­work Caffe, Baidu rec­om­mends that devel­op­ers use its Pad­dlePad­dle plat­form that was launched last year. Users can con­vert their ex­ist­ing Pad­dlePad­dle model to an MDL project us­ing sim­ple Python skills. Baidu has also given sup­port for projects such as Mo­bileNet, SqueezeNet and GoogLeNet.

MDL is avail­able on GitHub un­der the MIT li­cence.

Newspapers in English

Newspapers from India

© PressReader. All rights reserved.