Our company is actively hiring for a Computer Vision Algorithm Engineer position in Shenzhen, offering exceptional employment opportunities in humanoid robot perception technology. We are looking for talented professionals to develop real-time visual detection and semantic segmentation algorithms for humanoid robot scene understanding. This position provides the chance to work on cutting-edge VSLAM systems and build lightweight semantic maps for navigation systems. If you're searching for impactful careers in robotics perception, this represents one of the most exciting job openings in the humanoid robot industry.

The ideal candidate should meet the following requirements:

1. Proficient in semantic/instance segmentation network architecture (DeepLabV3+, Mask2Former, SAM, yolo adaptation model), master multi-scale feature fusion and attention mechanism optimization skills.
2. Familiar with small sample/weakly supervised segmentation schemes (such as Mask-CLIP, FocalClick).
3. In-depth understanding of the 2D/3D detection framework (DETR series, YOLO-v8/v9, PointPillars), and master the optimization methods of rotating frames and intensive scenes.
4. Be familiar with missed detection/false detection suppression strategies (such as timing consistency filtering) in dynamic scenarios.
5. Master the whole process of model training: data enhancement (CutMix/MixUp), loss function design (Focal Loss, Dice Loss), evaluation indicators (mAP, mIoU, PQ).
6. Proficient in using Python and CV libraries (OpenCV, PIL), you must master PyTorch (ability to customize layer/Dataset).

Extra points：
1. Be familiar with near-earth perspective data sets (such as HOMERobot, Habitat-Matterport 3D) or self-built simulation environments.
2. Study the coupling problem of binaural movement and perception: the influence of gait phase on image quality and compensation algorithm.
3. Master RGB-D point cloud segmentation (PointRend, PointGroup) or monocular depth estimation (AdaBins, ZoeDepth).
4. Develop geometric-assisted detection of multiple viewing angles (such as the use of motion parallax to improve the detection of small objects).
5. The visual basic model (VLMs) has been used to improve the generalization ability of zero samples.
6. Those who have top papers are preferred.

Check out the full job listings here