Raw Instinct: Trust Your Classifiers and Skip the Conversion

TL;DR

It is shown that a sufficiently advanced classifier can yield equivalent results on RAW input compared to RGB and a new public dataset is presented consisting of RAW images and the corresponding converted RGB images, confirming that classification performance can indeed be preserved.

Abstract

Using RAW-images in computer vision problems is surprisingly underexplored considering that converting from RAW to RGB does not introduce any new capture information. In this paper, we show that a sufficiently advanced classifier can yield equivalent results on RAW input compared to RGB and present a new public dataset consisting of RAW images and the corresponding converted RGB images. Classifying images directly from RAW is attractive, as it allows for skipping the conversion to RGB, lowering computation time significantly. Two CNN classifiers are used to classify the images in both formats, confirming that classification performance can indeed be preserved. We furthermore show that the total computation time from RAW image data to classification results for RAW images can be up to 8.46 times faster than RGB. These results contribute to the evidence found in related works, that using RAW images as direct input to computer vision algorithms looks very promising.

Figures (4)

• Figure 1: Simplified overview of the processes involved in RAW to RGB image conversion (Adapted from RAWtoRGB).
• Figure 2: Overview of the BCA implementation. The white blocks are convolutional blocks, the gray block is a downscaling block and the dark block is the CNN classifier. Figure inspired by BCAPaper.
• Figure 3: The 5 classes in the data capture: (a) arborio, (b) basmati, (c) brown, (d) jasmine and (e) parboiled.
• Figure 4: Front and top views of data capture.