A Deeply Quantized Classifier for Very Low Resolution ToF Imaging

This paper presents the design of tiny deeply quantized neural networks suitable for super integration within the Time-of-Flight, low-resolution, image sensor. They were aimed to achieve ultra low complexity with adequate classification accuracy. First floating-point models were studied to process 8-bits images 2 \(\times \) 2 pixel resolution, down-sampled from 8 \(\times \) 8 images of a public dataset. Data were acquired with an off-the-shelf Time-of-Flight sensor. Next, many deeply quantized networks were designed from scratch by using QKeras quantization training-aware schema. 8 and 6 bits pixel depth were generated by the sensor. Ternary, 6, and 8 bits quantizers were used along with convolutions and dense layers. Experimental results have shown that the proposed deeply quantized models achieved in maximum an accuracy of 77.79% compared to 88.48% for the floating-point models. The model size of the latters ranged from 92,320 bytes for the largest floating-point model to 151 bytes for the tiniest 6-bits deeply quantized model.