Keypoint detection is one of the main tasks of computer vision. Keypoints of objects can give us an insight into object characteristics, for example, human pose. Keypoint detection is a fast-growing research area and models are getting better and better. We have developed a model for keypoint detection on articulate objects. Those objects consist of two or more solid parts connected by joints that enable the rotation or translation of these objects. The model is developed for keypoint detection on gilthead seabream fish. We need to detect 18 keypoints of the fish. The locations of keypoints give us an insight into their morphological characteristics. Those characteristics can help us discover information about the habitat and way of feeding. The model we developed is based on the SwiftNet model with ResNet backbone with some modifications. We train the model on the dataset which consists of pictures of gilthead seabream and information about the keypoints. We are generating heatmaps from the keypoints given in the dataset. The heatmap is generated for each image. The number of channels in heatmaps is equal to the number of keypoints. We are validating the model parameters and data augmentations. We have developed a model and data augmentations that achieve an average distance between predicted and ground truth keypoints of 2.5 px on the test dataset. Taking into account the annotation error and the magnitude of the average distance, we can conclude that the results are satisfying.