IFE-CMT: Instance-Aware Fine-Grained Feature Enhancement Cross Modal Transformer for 3D Object Detection.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Multi-modal 3D object detection algorithms have advanced significantly.
• Current methods often overlook fine-grained features, impacting small object detection accuracy.
• Existing fusion strategies can lead to a decline in performance for detecting smaller objects.

Purpose of the Study:

• To propose a novel model, the Instance-aware Fine-grained feature Enhancement Cross Modal Transformer (IFE-CMT), for improved multi-modal 3D object detection.
• To enhance the detection accuracy of small objects within complex scenes.
• To address the limitations of current fusion strategies in capturing fine-grained object representations.

Main Methods:

• Developed an Instance feature Enhancement Module (IE-Module) for accurate multi-modal feature extraction and enhancement.
• Introduced a new point cloud branch network to expand the receptive field and improve semantic expression.
• Designed a cross-modal transformer architecture focusing on fine-grained feature enhancement.

Main Results:

• The IFE-CMT model demonstrated improved performance on the nuScenes dataset compared to the CMT model.
• Achieved a 2.1% and 1.9% increase in mAP on the validation and test sets, respectively.
• Significantly improved mAP for small objects like bicycles (6.6%) and motorcycles (3.7%).

Conclusions:

• The proposed IFE-CMT model effectively enhances fine-grained feature representations for multi-modal 3D object detection.
• IFE-CMT shows superior performance, particularly in detecting small objects, outperforming existing methods.
• The model offers a promising approach to address the challenges in small object detection within autonomous driving systems.