Vision processing unit: Difference between revisions

Revision as of 18:22, 1 May 2016 editArjayay (talk \| contribs)Autopatrolled, Extended confirmed users, Page movers, Pending changes reviewers, Rollbackers629,802 editsm Sp - Refered > Referred← Previous edit		Revision as of 22:58, 1 May 2016 edit undo81.153.147.173 (talk)No edit summaryNext edit →
Line 7:		Line 7:
	They are distinct from ]s, which contain specialised hardware for ] and ] (for 3D graphics), and whose ] is optimised for manipulating ] in ] ( reading ], and modifying ], with random access patterns).		They are distinct from ]s, which contain specialised hardware for ] and ] (for 3D graphics), and whose ] is optimised for manipulating ] in ] ( reading ], and modifying ], with random access patterns).

	Target markets are ], new classes of cameras for ] and ], and integrating machine vision acceleration into ]s and other ].		Target markets are ],], new classes of cameras for ] and ], and integrating machine vision acceleration into ]s and other ].

	== Examples ==		== Examples ==

Revision as of 22:58, 1 May 2016

A vision processing unit is an emerging class of microprocessor (as of 2016) designed to accelerate machine vision tasks.

These are distinct from video processing units (which are specialised for video encoding and decoding) in their suitability for running algorithms such as convolutional neural networks, SIFT etc.

They may include direct interfaces to take data from cameras (bypassing any off chip buffers), and have a greater emphasis on on-chip dataflow between many parallel execution units with scratchpad memory, like a manycore DSP. But, like video processing units, they may have a focus on low precision fixed point arithmetic for image processing.

They are distinct from GPUs, which contain specialised hardware for rasterization and texture mapping (for 3D graphics), and whose memory architecture is optimised for manipulating bitmap images in off-chip memory ( reading textures, and modifying frame buffers, with random access patterns).

Target markets are robotics,IoT, new classes of cameras for virtual reality and augmented reality, and integrating machine vision acceleration into smartphones and other mobile devices.

Movidius Myriad 2, which finds use in Google Project Tango.
Eyeriss, a design from MIT intended for running convolutional neural networks.
Microsoft hololens includes an accelerator referred to as a Holographic Processing Unit (complimentary to it's CPU and GPU), aimed at interpreting camera inputs, to accelerate environment tracking & vision for augmented reality applications.