| Title |
Service Platform for Serving Line-art Automatic Colorization Model |
| Authors |
이영섭(Yeongseop Lee) ; 이성진(Seongjin Lee) |
| DOI |
https://doi.org/10.5370/KIEEP.2022.71.1.41 |
| Keywords |
Machine Learning; Generative Adversarial Network; Line Arts Colorization; Image Generation |
| Abstract |
In this paper, we propose a service platform that can test and serve automatic colorization neural network models for labor-intensive colorization tasks using Generative Adversarial Networks (GANs). This service platform uses a model using a generator using two generators, a line loss function to increase the line data generalization ability of the model in the learning process, and data augmentation techniques to solve the line overfitting problem. This service efficiently supports inference on the CPU using ONNX (Open Neural Network Exchange) and serves as an inference server with a higher-order function-based pre-processor to support input/output of many sizes and a service front end for user and hint input. To test the inference performance of ONNX and torchscript. inference times were compared. Inference using the proposed ONNX averaged 0.4040, which was more than 5 times faster than 2.2683 using torchscript, enabling efficient inference. To test the inference performance of ONNX and torchscript, we checked inference times were compared. Inference using the proposed ONNX averaged 0.4040sec, which was more than 5 times faster than 2.2683sec using torchscript, enabling efficient inference. |