A face detection application on the CNP system was demonstrated achieving 10 fps on VGA images.

True if the cited work ([52], for example Farabet et al. on the CNP FPGA system) reports a face detection application demonstrated on the CNP that runs at about 10 frames per second on VGA (640x480) images. Mostly true if it was demonstrated on the CNP with a small difference in frame rate (roughly 8–12 fps) or a slightly different resolution. Mixed or mostly false if the frame rate or resolution differs substantially, for example 10 fps only at lower resolution, or if the demonstration was a simulation rather than on the system. False if no face detection demonstration on the CNP was reported.

Set before the research began.

1. E1 An FPGA-Based Stream Processor for Embedded Real-Time Vision with Convolutional Networks Farabet, Poulet, LeCun (Courant Institute, NYU); hosted on yann.lecun.com Own record

   “Real-time face detection is demonstrated, with speeds of 10 frames per second at VGA resolution.”

2. E2 An FPGA-Based Stream Processor for Embedded Real-Time Vision with Convolutional Networks Farabet, Poulet, LeCun (Courant Institute, NYU); hosted on yann.lecun.com Own record

   “Computing the same ConvNet on a monoscale input reduces this time by a factor of 1.5, or allows full VGA frames (640 × 480) to be processed at 10fps.”

3. E3 CNP: an fpga-based processor for convolutional networks Farabet, Poulet, Han, LeCun (Courant Institute, NYU; Perceptive Pixel Inc.); hosted on yann.lecun.com · 2009 Own record

   “Face detection on a 512 × 384 frame takes 100ms (10 frames per second), which corresponds to an average performance of 3.4×109 connections per second for this 340 million connection network.”

4. E4 FeatherNet: An Accelerated Convolutional Neural Network Design for Resource-constrained FPGAs ACM Transactions on Reconfigurable Technology and Systems (authors at American University of Beirut and Intel) · March 2019

   “The CNP could process 10 frames per second using a ConvNet with 2 million synaptic connections.”

The abstract of the CNP paper by Farabet, Poulet and LeCun says that real-time face detection was demonstrated at 10 frames per second at VGA resolution (E1). The same paper adds that running the ConvNet on a single-scale (monoscale) input allows full VGA frames (640×480) to be processed at 10 fps (E2). The companion FPL 2009 CNP paper reports face detection at 10 fps on 512×384 frames (E3), a somewhat lower resolution than VGA. A later independent review confirms that the CNP processed 10 frames per second but names no resolution or application (E4), so the claim holds overall with a small qualification about resolution.