SRC-N ‘s published paper “Automatic Detection and Classification System of Domestic Waste via Multi-Model Cascaded Convolutional Neural Network” and “Self-Aligned Dual Face Regression Networks for Robust 3D Dense Face Alignment and Reconstruction” was accepted. And published in IEEE Journal.

1. “Automatic Detection and Classification System of Domestic Waste via Multi-Model Cascaded Convolutional Neural Network”: This paper has been published in IEEE Transactions on Industrial Informatics, the paper proposes a MCCNN architecture and a STC system based on MCCNN Experimental results showed a state-of-the-art performance, with an average improvement of 10% in detection precision.

Figure 1 shows the architecture of MCCNN (Multi-model Cascaded Convolutional Neural Network) for domestic waste image detection. MCCNN combined three subnetworks to obtain the detections. Moreover, to suppress the false-positive predicts, we utilized a classification model cascaded with the detection part to judge whether the detection results are correct.

Figure 2 shows the pipeline of waste disposal process. The STC (Smart Trach Can) system serves as a front-end part of the entire waste disposal process. The system takes an image of waste through the camera and uses the MCCNN model to recognize the image. The recognition results will be displayed on the screen. After that, the system automatically opens the trays, and the waste falls into the trash can.

2. “Automatic Detection and Classification System of Domestic Waste via Multi-Model Cascaded Convolutional Neural Network”: This paper has been published in IEEE Transactions on Image Processing. The paper proposes a self-aligned dual face regression network (SADRNet) to solve the problem of 3D face reconstruction and dense alignment under unconstrained conditions. Extensive experiments on two popular benchmarks, AFLW2000-3D and Florence, demonstrate that the proposed method achieves significant superior performance over existing state-of-the-art methods.

Figure 1 shows the architecture of proposed self-aligned dual face regression network (SADRNet). A is the attention mask. P is the pose-dependent face. D is the face shape deformation (visualized in UV space). S is the mean face template. S is the pose-independent face. G is the output face model.
Figure 2 shows some results on AFLW2000-3D from our algorithm. Our results are more accurate than the ground truth. From the top row to the bottom row are the input images, the sparse alignment results of SADRNet and the corresponding ground truth (blue for our method and red for the ground truth), the reconstructed face models, and the ground truth face models.