A 345mW heterogeneous many-core processor with an intelligent inference engine for robust object recognition
A 228GOPS 345 mW heterogeneous many-core processor combines bottom-up and top-down attention for ... more A 228GOPS 345 mW heterogeneous many-core processor combines bottom-up and top-down attention for high accuracy. It uses analog-digital mixed-mode neuro-fuzzy inference circuits to realize robust object recognition. Weight perturbation learning and workload-aware voltage/frequency control are adopted. The 5 Ã 10 mm2 chip with 96% recognition accuracy contains 4 SIMD vector processing elements and 32 MIMD processors.
The visual attention mechanism, which is the way humans perform object recognition, was applied t... more The visual attention mechanism, which is the way humans perform object recognition, was applied to the implementation of a high performance object recognition chip. Even though the previous chip achieved 50% gain of computational cost, it could recognize only one object in a frame so that it is not suitable for advanced multi-object recognition applications such as video surveillance, intelligent robots, and autonomous vehicle
The Visual Attention Engine(VAE), an 80x60 digital Cellular Neural Network, rapidly extracts glob... more The Visual Attention Engine(VAE), an 80x60 digital Cellular Neural Network, rapidly extracts global features used as attentional cues to streamline detailed object recognition. A peak performance of 24GOPS is achieved by 120 processing elements (PE) shared by the cells. 2D Shift register based data transactions enable 93% PE utilization. Integrated within an object recognition SoC, the 4.5mm 2 VAE running at 200MHz improves object recognition frame rate by 83% while consuming just 84mW.
Even though visual attention models using bottom-up saliency can speed up object recognition by p... more Even though visual attention models using bottom-up saliency can speed up object recognition by predicting object locations, in the presence of multiple salient objects, saliency alone cannot discern target objects from the clutter in a scene. Using a metric named familiarity, we propose a top-down method for guiding attention towards target objects, in addition to bottom-up saliency. To demonstrate the effectiveness of familiarity, the unified visual attention model (UVAM) which combines top-down familiarity and bottom-up saliency is applied to SIFT based object recognition. The UVAM is tested on 3600 artificially generated images containing COIL-100 objects with varying amounts of clutter, and on 126 images of real scenes. The recognition times are reduced by 2.7× and 2×, respectively, with no reduction in recognition accuracy, demonstrating the effectiveness and robustness of the familiarity based UVAM.
A 125GOPS 583mW Network-on-Chip Based Parallel Processor with Bio-inspired Visual-Attention Engine
AbstractA network-on-chip (NoC) based parallel processor is presented for bio-inspired real-time... more AbstractA network-on-chip (NoC) based parallel processor is presented for bio-inspired real-time object recognition with visual attention algorithm. It contains an ARM10-compatible 32-bit main processor, 8 single-instruction multiple-data (SIMD) clusters with 8 processing ...
The visual attention mechanism, which is the way humans perform object recognition , was applied ... more The visual attention mechanism, which is the way humans perform object recognition , was applied to the implementation of a high performance object recognition chip . Even though the previous chip achieved 50% gain of computational cost [2], it could recognize only one object in a frame so that it is not suitable for advanced multi-object recognition applications such as video surveillance, intelligent robots, and autonomous vehicle navigation .
A 345mW heterogeneous many-core processor with an intelligent inference engine for robust object recognition
A 228GOPS 345 mW heterogeneous many-core processor combines bottom-up and top-down attention for ... more A 228GOPS 345 mW heterogeneous many-core processor combines bottom-up and top-down attention for high accuracy. It uses analog-digital mixed-mode neuro-fuzzy inference circuits to realize robust object recognition. Weight perturbation learning and workload-aware voltage/frequency control are adopted. The 5 Ã 10 mm2 chip with 96% recognition accuracy contains 4 SIMD vector processing elements and 32 MIMD processors.
The visual attention mechanism, which is the way humans perform object recognition, was applied t... more The visual attention mechanism, which is the way humans perform object recognition, was applied to the implementation of a high performance object recognition chip. Even though the previous chip achieved 50% gain of computational cost, it could recognize only one object in a frame so that it is not suitable for advanced multi-object recognition applications such as video surveillance, intelligent robots, and autonomous vehicle
The Visual Attention Engine(VAE), an 80x60 digital Cellular Neural Network, rapidly extracts glob... more The Visual Attention Engine(VAE), an 80x60 digital Cellular Neural Network, rapidly extracts global features used as attentional cues to streamline detailed object recognition. A peak performance of 24GOPS is achieved by 120 processing elements (PE) shared by the cells. 2D Shift register based data transactions enable 93% PE utilization. Integrated within an object recognition SoC, the 4.5mm 2 VAE running at 200MHz improves object recognition frame rate by 83% while consuming just 84mW.
Even though visual attention models using bottom-up saliency can speed up object recognition by p... more Even though visual attention models using bottom-up saliency can speed up object recognition by predicting object locations, in the presence of multiple salient objects, saliency alone cannot discern target objects from the clutter in a scene. Using a metric named familiarity, we propose a top-down method for guiding attention towards target objects, in addition to bottom-up saliency. To demonstrate the effectiveness of familiarity, the unified visual attention model (UVAM) which combines top-down familiarity and bottom-up saliency is applied to SIFT based object recognition. The UVAM is tested on 3600 artificially generated images containing COIL-100 objects with varying amounts of clutter, and on 126 images of real scenes. The recognition times are reduced by 2.7× and 2×, respectively, with no reduction in recognition accuracy, demonstrating the effectiveness and robustness of the familiarity based UVAM.
A 125GOPS 583mW Network-on-Chip Based Parallel Processor with Bio-inspired Visual-Attention Engine
AbstractA network-on-chip (NoC) based parallel processor is presented for bio-inspired real-time... more AbstractA network-on-chip (NoC) based parallel processor is presented for bio-inspired real-time object recognition with visual attention algorithm. It contains an ARM10-compatible 32-bit main processor, 8 single-instruction multiple-data (SIMD) clusters with 8 processing ...
The visual attention mechanism, which is the way humans perform object recognition , was applied ... more The visual attention mechanism, which is the way humans perform object recognition , was applied to the implementation of a high performance object recognition chip . Even though the previous chip achieved 50% gain of computational cost [2], it could recognize only one object in a frame so that it is not suitable for advanced multi-object recognition applications such as video surveillance, intelligent robots, and autonomous vehicle navigation .
Uploads
Papers by Minsu Kim