Zhidong Liang

Zehan Zhang, Ming Zhang, Zhidong Liang et al.

3D vehicle detection based on multi-modal fusion is an important task of many applications such as autonomous driving. Although significant progress has been made, we still observe two aspects that need to be further improvement: First, the specific gain that camera images can bring to 3D detection is seldom explored by previous works. Second, many fusion algorithms run slowly, which is essential for applications with high real-time requirements(autonomous driving). To this end, we propose an end-to-end trainable single-stage multi-modal feature adaptive network in this paper, which uses image information to effectively reduce false positive of 3D detection and has a fast detection speed. A multi-modal adaptive feature fusion module based on channel attention mechanism is proposed to enable the network to adaptively use the feature of each modal. Based on the above mechanism, two fusion technologies are proposed to adapt to different usage scenarios: PointAttentionFusion is suitable for filtering simple false positive and faster; DenseAttentionFusion is suitable for filtering more difficult false positive and has better overall performance. Experimental results on the KITTI dataset demonstrate significant improvement in filtering false positive over the approach using only point cloud data. Furthermore, the proposed method can provide competitive results and has the fastest speed compared to the published state-of-the-art multi-modal methods in the KITTI benchmark.

Zhidong Liang, Ming Zhang, Zehan Zhang et al.

We present RangeRCNN, a novel and effective 3D object detection framework based on the range image representation. Most existing methods are voxel-based or point-based. Though several optimizations have been introduced to ease the sparsity issue and speed up the running time, the two representations are still computationally inefficient. Compared to them, the range image representation is dense and compact which can exploit powerful 2D convolution. Even so, the range image is not preferred in 3D object detection due to scale variation and occlusion. In this paper, we utilize the dilated residual block (DRB) to better adapt different object scales and obtain a more flexible receptive field. Considering scale variation and occlusion, we propose the RV-PV-BEV (range view-point view-bird's eye view) module to transfer features from RV to BEV. The anchor is defined in BEV which avoids scale variation and occlusion. Neither RV nor BEV can provide enough information for height estimation; therefore, we propose a two-stage RCNN for better 3D detection performance. The aforementioned point view not only serves as a bridge from RV to BEV but also provides pointwise features for RCNN. Experiments show that RangeRCNN achieves state-of-the-art performance on the KITTI dataset and the Waymo Open dataset, and provides more possibilities for real-time 3D object detection. We further introduce and discuss the data augmentation strategy for the range image based method, which will be very valuable for future research on range image.

Zhidong Liang, Ming Yang, Chunxiang Wang

This paper introduces a novel approach for 3D semantic instance segmentation on point clouds. A 3D convolutional neural network called submanifold sparse convolutional network is used to generate semantic predictions and instance embeddings simultaneously. To obtain discriminative embeddings for each 3D instance, a structure-aware loss function is proposed which considers both the structure information and the embedding information. To get more consistent embeddings for each 3D instance, attention-based k nearest neighbour (KNN) is proposed to assign different weights for different neighbours. Based on the attention-based KNN, we add a graph convolutional network after the sparse convolutional network to get refined embeddings. Our network can be trained end-to-end. A simple mean-shift algorithm is utilized to cluster refined embeddings to get final instance predictions. As a result, our framework can output both the semantic prediction and the instance prediction. Experiments show that our approach outperforms all state-of-art methods on ScanNet benchmark and NYUv2 dataset.