Rotated object detectors commonly encounter instability during the training process, primarily due to background noise and angular periodicity. Targets with elongated or non-convex shapes may introduce background noise during convolution, hindering accurately extracting features. Meanwhile, the periodicity of angles leads to predictions beyond the defined range, subsequently impeding the convergence. This letter introduces an Angle Adaptive Module (AAM) designed for the backbone, enhancing the ability of the model to accurately extract object features and dynamically select the optimal angle. Moreover, to mitigate the effect of angle periodicity, a method called Radian Regression Method (RRM) is proposed for predicting proper angles. It avoids directly regressing the value and instead produces the probability density distribution of the offset. We elaborately design numerous experiments to demonstrate the effectiveness of the proposed modules. As a result, the proposed method attains competitive results across various datasets, including DOTAv1.0, DOTAv1.5, DOTAv2.0, HRSC2016, and DIOR-R.

Visual grounding is a common vision task that involves grounding descriptive sentences to the corresponding regions of an image. Most existing methods use independent image-text encoding and apply complex hand-crafted modules or encoder-decoder architectures for modal interaction and query reasoning. However, their performance significantly drops when dealing with complex textual expressions. This is because the former paradigm only utilizes limited downstream data to fit the multi-modal feature fusion. Therefore, it is only effective when the textual expressions are relatively simple. In contrast, given the wide diversity of textual expressions and the uniqueness of downstream training data, the existing fusion module, which extracts multimodal content from a visual-linguistic context, has not been fully investigated. In this paper, we present a simple yet robust transformer-based framework, SimVG, for visual grounding. Specifically, we decouple visual-linguistic feature fusion from downstream tasks by leveraging existing multimodal pre-trained models and incorporating additional object tokens to facilitate deep integration of downstream and pre-training tasks. Furthermore, we design a dynamic weight-balance distillation method in the multi-branch synchronous learning process to enhance the representation capability of the simpler branch. This branch only consists of a lightweight MLP, which simplifies the structure and improves reasoning speed. Experiments on six widely used VG datasets, i.e., RefCOCO/+/g, ReferIt, Flickr30K, and GRefCOCO, demonstrate the superiority of SimVG. Finally, the proposed method not only achieves improvements in efficiency and convergence speed but also attains new state-of-the-art performance on these benchmarks. Codes and models are available at https://github.com/Dmmm1997/SimVG.

Due to the inherent mismatch between rotated objects and horizontal features, feature point misalignment has been a challenge in the task of Rotated Object Detection (ROD). Specifically, considering the pattern of convolution, foreground features are often mixed with background noise, which can confuse the model and affect the model from feature point alignment during the training phase. To mitigate this issue, previous methods concentrate on fixed positions derived from predicted boxes by additionally introducing a refinement stage. However, merely learning fixed position priors during training can result in suboptimal alignment and inefficiency during inference. This paper introduces a dynamic point alignment detector to concurrently address issues associated with feature misalignment and inference inefficiency. The method is made up of two components: the fine-grained points generator (FPG) captures key information, and the point alignment module (PAM) derives precise feature representations. Both modules empower the detector with the capability to dynamically perceive rotated objects, extracting more comprehensive and reasoned feature contents from the feature maps. In general, our method enables the model to independently identify and prioritize the valuable features during the training process. Subsequently, during the inference stage, the results can be directly predicted without additional alignment operations. The experimental results demonstrate that our method can achieve competitive and superior results with average precision (AP) values of 79.33%, 95.73%, and 63.37% on the DOTA, HRSC2016, and DIOR-R datasets, respectively. Codes will be publicly available at https://github.com/Xuyihaoby/PARDet.