Suren Kumar

US Patent:

20210150249, May 20, 2021

Filed:

Jan 28, 2021

Appl. No.:

17/160399

Inventors:

- Madison WI, US
Suren Kumar - Seattle WA, US

International Classification:

G06K 9/46
G06N 3/08
G06Q 30/06
G06K 9/32

Abstract:

Methods and systems for training a metric learning convolutional neural network (CNN)-based model for cross-domain image retrieval are disclosed. The methods and systems perform steps of generating a plurality of batches sampled from a cross-domain training dataset to train the CNN-based model to match images of different sub-categories from one domain to another, and training the CNN-based model using the generated batches. The CNN-based model comprises various pooling, normalization, and concatenation layers that enable it to concatenate the normalized outputs of multiple concatenation layers. Use of the generated batches comprises executing a loss function based on one or more batches, where the loss function is a triplet, contrastive, or cluster loss function. Embodiments of the present invention enable the CNN-based model to summarize information from multiple convolutional layers, thus improving visual search. Also disclosed are benefits of the new methods, and alternative embodiments of implementation.

US Patent:

20150324656, Nov 12, 2015

Filed:

May 9, 2014

Appl. No.:

14/274073

Inventors:

- Cambridge MA, US
Michael J. Jones - Belmont MA, US
Suren Kumar - Amherst NY, US

Assignee:

Mitsubishi Electric Research Laboratories, Inc. - Cambridge MA

International Classification:

G06K 9/46
G06K 9/52
H04N 7/18
H04N 5/33

Abstract:

A method and system tracks objects in an environment by acquiring a first sequence of images of the environment with a visible-light camera having a first frame rate and a second sequence of images with a thermal infrared sensor having a second frame rate. The second frame rate is substantially lower than the first frame rate. The objects are tracked in the first sequence of images to obtain tracks. Warm regions in the second sequence of images are detected to obtain detections. The tracks and the detections are aligned spatially and temporally, and verified to determine whether the tracks and detections coincide after the aligning.