An Improved YOLOX Algorithm based on Structural Re-parameterized CBAM for Wild Animals Detection

We use the structural re-parameterization method to optimize the generation process of the attention map. CBAM has long been among the most widely used attention techniques because of its simplicity and superior performance. The success of RepVgg shows great application potential of the structural re-parameterization methods, which we notice that could be used to further strengthen the ability of the CBAM. Based on YOLOX, we evaluate our method on a self-built wild animals' dataset. The result shows better performance compared to the original CBAM. Notably, the gain is FREE OF CHARGE. Compared to the baseline model, no extra parameters or FLOPs are introduced during the inference stage. Experiments prove that it is feasible to apply the idea of structural re-parameterization in the generation process of attention mechanism. The proposed method shows great application potential on wild animal detection task.