English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Tools annual

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Semantic segmentation has witnessed tremendous progress due to the proposalof various advanced network architectures. However, they are extremely hungryfor delicate annotations to train, and the acquisition is laborious andunaffordable. Therefore, we present FreeMask in this work, which resorts tosynthetic images from generative models to ease the burden of both datacollection and annotation procedures. Concretely, we first synthesize abundanttraining images conditioned on the semantic masks provided by realisticdatasets. This yields extra well-aligned image-mask training pairs for semanticsegmentation models. We surprisingly observe that, solely trained withsynthetic images, we already achieve comparable performance with real ones(e.g., 48.3 vs. 48.5 mIoU on ADE20K, and 49.3 vs. 50.5 on COCO-Stuff). Then, weinvestigate the role of synthetic images by joint training with real images, orpre-training for real images. Meantime, we design a robust filtering principleto suppress incorrectly synthesized regions. In addition, we propose toinequally treat different semantic masks to prioritize those harder ones andsample more corresponding synthetic images for them. As a result, eitherjointly trained or pre-trained with our filtered and re-sampled synthesizedimages, segmentation models can be greatly enhanced, e.g., from 48.7 to 52.0 onADE20K. Code is available at https://github.com/LiheYoung/FreeMask.

FreeMask: Synthetic Images with Dense Annotations Make Stronger Segmentation Models