Mixed Non-linear Quantization for Vision Transformers

The majority of quantization methods have been proposed to reduce the model size of Vision Transformers, yet most of them have overlooked the quantization of non-linear operations. Only a few works have addressed quantization for non-linear operations, but they applied a single quantization method across all non-linear operations. We believe that this can be further improved by employing a different quantization method for each non-linear operation. Therefore, to assign the most error-minimizing quantization method from the known methods to each non-linear layer, we propose a mixed non-linear quantization that considers layer-wise quantization sensitivity measured by SQNR difference metric. The results show that our method outperforms I-BERT, FQ-ViT, and I-ViT in both 8-bit and 6-bit settings for ViT, DeiT, and Swin models by an average of 0.6%p and 19.6%p, respectively. Our method outperforms I-BERT and I-ViT by 0.6%p and 20.8%p, respectively, when training time is limited. We plan to release our code at https://gitlab.com/ones-ai/mixed-non-linear-quantization.

Simplifying Two-Stage Detectors for On-Device Inference in Remote Sensing

Deep learning has been successfully applied to object detection from remotely sensed images. Images are typically processed on the ground rather than on-board due to the computation power of the ground system. Such offloaded processing causes delays in acquiring target mission information, which hinders its application to real-time use cases. For on-device object detection, researches have been conducted on designing efficient detectors or model compression to reduce inference latency. However, highly accurate two-stage detectors still need further exploitation for acceleration. In this paper, we propose a model simplification method for two-stage object detectors. Instead of constructing a general feature pyramid, we utilize only one feature extraction in the two-stage detector. To compensate for the accuracy drop, we apply a high pass filter to the RPN's score map. Our approach is applicable to any two-stage detector using a feature pyramid network. In the experiments with state-of-the-art two-stage detectors such as ReDet, Oriented-RCNN, and LSKNet, our method reduced computation costs upto 61.2% with the accuracy loss within 2.1% on the DOTAv1.5 dataset. Source code will be released.

DyRA: Dynamic Resolution Adjustment for Scale-robust Object Detection

In object detection, achieving constant accuracy is challenging due to the variability of object sizes. One possible solution to this problem is to optimize the input resolution, known as a multi-resolution strategy. Previous approaches for optimizing resolution are often based on pre-defined resolutions or a dynamic neural network, but there is a lack of study for run-time resolution optimization for existing architecture. In this paper, we propose an adaptive resolution scaling network called DyRA, which comprises convolutions and transformer encoder blocks, for existing detectors. Our DyRA returns a scale factor from an input image, which enables instance-specific scaling. This network is jointly trained with detectors with specially designed loss functions, namely ParetoScaleLoss and BalanceLoss. The ParetoScaleLoss produces an adaptive scale factor from the image, while the BalanceLoss optimizes the scale factor according to localization power for the dataset. The loss function is designed to minimize accuracy drop about the contrasting objective of small and large objects. Our experiments on COCO, RetinaNet, Faster-RCNN, FCOS, and Mask-RCNN achieved 1.3%, 1.1%, 1.3%, and 0.8% accuracy improvement than a multi-resolution baseline with solely resolution adjustment. The code is available at https://github.com/DaEunFullGrace/DyRA.git.

A Unified Compression Framework for Efficient Speech-Driven Talking-Face Generation

Virtual humans have gained considerable attention in numerous industries, e.g., entertainment and e-commerce. As a core technology, synthesizing photorealistic face frames from target speech and facial identity has been actively studied with generative adversarial networks. Despite remarkable results of modern talking-face generation models, they often entail high computational burdens, which limit their efficient deployment. This study aims to develop a lightweight model for speech-driven talking-face synthesis. We build a compact generator by removing the residual blocks and reducing the channel width from Wav2Lip, a popular talking-face generator. We also present a knowledge distillation scheme to stably yet effectively train the small-capacity generator without adversarial learning. We reduce the number of parameters and MACs by 28$\times$ while retaining the performance of the original model. Moreover, to alleviate a severe performance drop when converting the whole generator to INT8 precision, we adopt a selective quantization method that uses FP16 for the quantization-sensitive layers and INT8 for the other layers. Using this mixed precision, we achieve up to a 19$\times$ speedup on edge GPUs without noticeably compromising the generation quality.