illustration of a green net
2021年10月11日 - 2021年10月17日

Microsoft at ICCV 2021

地点： Virtual

微软研究院博客

On the left, a diagram with three layers, each of which contains a half-transparent image processed from the same image. The processed images are partitioned into several grids, and each grid contains 4 x 4 image patches. From bottom to top, the number of grids in each layer are 4 x 4, 2 x 2, and 1 x 1, respectively. The layers are labeled “4x,” “8x,” and “16x,” respectively, from bottom to top. An arrow joining the three layers points upward to the words “Segmentation” and “Detection” and an ellipsis. Another arrow points from the top layer to the word “classification.” On the right, a bar chart with a blue bar labeled “Swin V1” and an orange bar labeled “Swin V2.” The orange bar is much taller and labeled “3 billion (1,536 x 1,536 resolution)”; the blue bar is labeled “197 million.” An arrow labeled “15x” points upward from the blue bar, indicating the orange bar is 15 times higher than the blue one.
微软研究院博客

| Han Hu 和 Baining Guo

Early last year, our research team from the Visual Computing Group (opens in new tab) introduced Swin Transformer (opens in new tab), a Transformer-based general-purpose computer vision architecture that for the first time beat convolutional neural networks on the important…

An image generation model training using MVM. Images show a panda progressively being generated to become more realistic, various paintings generated by the model, an anime character generated to become more realistic, and various cat images generated to become more realistic.
微软研究院博客

| Mengyu Dai 和 Junwon Park

Generative image models offer a unique value by creating new images. Such images can be sharp super-resolution versions of existing images or even realistic-looking synthetic photographs. Generative Adversarial Networks (GANs) and their variants have demonstrated pioneering success with the framework…

ACAV100M text on top of a series of small images.
微软研究院博客

| Yale Song

The natural association between visual observations and their corresponding sounds has exhibited powerful self-supervision signals for learning video representations (opens in new tab), which makes the ever-growing amount of online video an attractive data source for self-supervised learning. However, online…

On left, text reads ORBIT benchmark dataset: 77 blind and low-vision collectors, 486 objects, 3822 videos, and 2687934 frames. On right, a graphic of a face mask with a line that connects to a picture of a cloth mask with a black and white zig-zag pattern. The line reads seven to eight videos per object. Below the face mask graphic, there are three yellow objects resembling a watering can, a key, and a comb. A line next to these reads two to ten objects per user. The objects are falling into a green bucket, with a line to the right of the bucket that reads user’s bucket.
微软研究院博客

| Daniela Massiceti, Cecily Morrison, Katja Hofmann, 和 Ed Cutrell

Object recognition systems have made spectacular advances in recent years, but they rely on training datasets with thousands of high-quality, labelled examples per object category. Learning new objects from only a few examples could open the door to many new…