Many surveys and reviews like [22, 23, 39] describe semantic segmentation as the Computer Vision (CV) task of predicting a category label at the pixel level. It builds upon simpler vision tasks such as image classification and object detection, and also shares some similarities with more advanced challenges like parts segmentation, instance segmentation, and panoptic segmentation. A visual comparison between the related Computer Vision (CV) tasks is reported in Figure 1. Image classification aims at understanding the overall scene in an image by giving it one or more labels, while object detection (Figure 1b) focuses on predicting the location of one or more objects in an image usually providing bounding boxes. Pixel-level prediction tasks like parts segmentation (Figure 1d) is a closer problem to semantic segmentation (Figure 1c), as it aims at predicting pixel-level segmentation masks covering the parts that compose the intended subject, such as face parts like the chin, nose and eyes. Instance segmentation (Figure 1e) aims to distinguish individual objects in an image, even if they are of the same kind, but does not necessarily assign them a category. Finally, panoptic segmentation (Figure 1f) combines semantic segmentation with instance segmentation, predicting the pixel-level category and distinguishing each object in the scene. Overall, we can place semantic segmentation as a midpoint on a spectrum of image understanding tasks ranging from coarse to fine.

Copy-Paste has proven to be a very effective data augmentation for instance segmentation which can improve the generalization of the model. We used a task-specific Copy-Paste data augmentation method to achieve good performance on the instance segmentation track of the 2nd VIPriors workshop challenge. We also applied additional data augmentation techniques including RandAugment and GridMask. Our segmentation model is the HTC detector on the CBSwin-B with CBFPN with some tweaks. This model was trained at the multi-scale mode by a random sampler on the 6x schedule and tested at the single-scale mode. By combining these techniques, we achieved 0.398 AP@0.50:0.95 with the validation set and 0.433 AP@0.50:0.95 with the test set. Finally, we reached 0.477 AP@0.50:0.95 with the test set by adding the validation set to the training data. Source code is available at https://github.com/jahongir7174/VIP2021.