MIT researchers argue that the ARC benchmark is essentially a visual problem and present the Vision ARC (VARC) framework, which reformulates ARC as an image‑to‑image translation task using a Vision Transformer, achieving human‑level accuracy through a novel canvas representation and test‑time training.
This article provides an in‑depth overview of Generative Adversarial Networks (GANs), covering their original formulation, major variants such as DCGAN and WGAN, challenges like mode collapse, image‑to‑image translation techniques (cGAN, pix2pix, CycleGAN), and practical handwriting style‑transfer implementations using BicycleGAN and Zi2Zi.
This paper presents a scale‑space‑based deep neural network that treats intrinsic image decomposition as an image‑to‑image translation, introduces a multi‑channel Laplacian pyramid architecture with novel loss and data‑augmentation strategies, and demonstrates superior performance on MPI‑Sintel and MIT Intrinsic datasets.
Researchers propose an unsupervised GAN-based image-to-image translation method that automatically dresses nude women in bikinis, preserving semantic content while removing sensitive parts, using unpaired datasets and Mask‑RCNN background removal, demonstrating impressive visual results without manual annotation.
