Shen Zheng 「郑深」

I am a second-year PhD student at the Robotics Institute, Carnegie Mellon University (CMU), advised by Dr. Srinivasa Narasimhan. I have interned at Waymo and Momenta, and worked at Lucid Motors. I completed my M.S. in Computer Vision (MSCV) at CMU, and earned my B.S. in Mathematics from Wenzhou-Kean University (WKU), where I worked with Dr. Gaurav Gupta.

Email1: [email protected]

Email2: [email protected]

Resume  /  Google Scholar  /  Github  /  Leetcode

My current research takes a data-centric approach to enable robust understanding of long-tail yet safety-critical scenarios in driving (e.g., low-light, bad weather, work zones):

• Real-world long-tail collection (e.g., ROADWork)
• Contextual background generation (e.g., TPSeNCE)
• Sharp foreground generation (e.g., Instance-Warp)
• Historical-to-future prediction (WIP on World Models)

My earlier works focused on image restoration and enhancement (e.g., SGZ, LLIE_Survey).

Motivation: Navigating through work zones is challenging due to a lack of large-scale open datasets.

Solution: Introduce the ROADWork dataset, which is so far the largest open-source work zone dataset, to help learn how to recognize, observe, analyze, and drive through work zones.

Motivation: Domain adaptation methods struggle to learn smaller objects amidst dominant backgrounds with high cross-domain variations.

Solution: Warp source-domain images in-place using instance-level saliency to oversample objects and undersample backgrounds during domain adaptation training.

Motivation: Previous image-to-image translation methods produce artifacts and distortions, and lack control over the amount of rain generated.

Solution: Introduce a Triangular Probability Similarity (TPS) loss to minimize the artifacts and distortions during rain generation. Propose a Semantic Noise Contrastive Estimation (SeNCE) strategy to optimize the amounts of generated rain. Show that realistic rain generation benefits deraining and object detection in rain.

Motivation: Existing LLIE datasets focus on either overexposure or underexposure, not both, and usually feature minimally degraded images captured from static positions.

Solution: Present a comprehensive survey of low-light image enhancement (LLIE). Propose the SICE_Grad and SICE_Mix image datasets, which include images with both overexposure and underexposure. Introduce Night Wenzhou, a large-scale, high-resolution video dataset captured in fast motion with diverse illuminations and degradation.

Motivation: Current point cloud analysis methods struggle with irregular (i.e., unevenly distributed) point clouds.

Solution: PointNorm, a point cloud analysis network with a DualNorm module (Point Normalization & Reverse Point Normalization) that leverages local mean and global standard deviation.

Motivation: Current low-light image enhancement methods cannot handle uneven illuminations, is computationally inefficient, and fail to preserve the semantic information.

Solution: Introduce SGZ, a zero-shot low-light image enhancement framework with pixel-wise light deficiency estimation, parameter-free recurrent image enhancement, and unsupervised semantic segmentation.

Motivation: Generative models experience posterior collapse and vanishing gradient due to no effective metric for real-fake image evaluation.

Solution: Propose Adversarial Similarity Distance Introspective Variational Autoencoder (AS-IntroVAE), which can address the posterior collapse and the vanishing gradient problem in image generation in one go.

Mentor: Guohao Zhang

(WIP) Improved Online HD Map Construction using long-term and short-term memory fusion.

Director: Dr. Feng Guo

Working as a perception software engineer in the ADAS perception team responsible for ADAS parking, traffic light detection, and blockage detection.

Improved BEVFormer for ADAS parking (reverse & parallel) by using extrinsic calibration to interpolate and smooth edges to enhance curb detection.

Trained YOLO6 on full-resolution images containing traffic lights and fine-tuned arrow types, confidence, IoU, and area thresholds, resulting in a 40+% improvement in mAP (final mAP: 98%+ for day; 90%+ for night).

Developed a binary semantic segmentation model based on CenterNet to detect blockages such as ice, snow, mud, mud blur, rain drops, and sun glares, achieving 93%+ IoU.

Director: Dr. Wangjiang Zhu

Responsible for long-tailed data augmentation, training data auto-labeling and cleaning, and model evaluation for traffic light detection algorithms.

Implemented CycleGAN to conduct unsupervised data augmentation, converting traffic light bulbs from left arrow to round & leftUturn arrow.

Constructed a traffic light auto-label model using quantized VoVNet-57, filtering 14,618 incorrect annotations from 1,160,513 labeled frames.

Increased the classification accuracy for leftUturn traffic light from 78.41% to 87.27%, and the mean average precision from 93.01% to 94.80%.