Diabetic foot ulcers (DFUs) are a significant condition where early detection is crucial due to the steadily increasing global prevalence of affected patients. In this paper, we aim to assist in the early detection and treatment of DFU by performing instance segmentation on DFU images using a self-supervised method. Using the pre-trained weights of Dinov2, which is based on a self-supervised learning approach, we connected a Mask R-CNN model to Dinov2’s head for instance segmentation. We divide the images provided by the DFU challenge into training and validation sets, with data augmentation applied to enhance performance. The performance of the model was evaluated using Dice score and IoU metrics. On our generated validation dataset, which consists of 400 images, we achieved a maximum Dice score of 0.8181. On the leaderboard, where performance was measured using the Test set provided by DFUC2024, the highest Dice score was 0.6764. By leveraging the self-supervised learning backbone to capture intrinsic features of the images, we achieved superior performance compared to traditional instance segmentation methods. The model and code we have implemented are publicly available at https://github.com/seihyoung/dfuc2024 .