xuexigua Goto Github PK

ancient-hand-writtten-recognition-under-data-deficiency

Ancient Indian Languages OCR: Enhancing handwritten character recognition using Capsule Networks with minimal data - A novel approach leveraging advanced CapsNets to interpret historical texts accurately with only 190 samples per character, addressing the challenge of scarce labeled datasets in preserving cultural heritage.

cameracalibration

Fisheye or Normal Camera Intrinsic and Extrinsic Calibration. Surround Camera Bird Eye View Generator.

cplusplusthings

C++那些事

cross_view_transformers

Cross-view Transformers for real-time Map-view Semantic Segmentation (CVPR 2022 Oral)

cvpr2021_pfnet

cvpr2022-paper-code-interpretation

cvpr2022/cvpr2021/cvpr2020/cvpr2019/cvpr2018/cvpr2017 论文/代码/解读/直播合集，极市团队整理

deep-learning-interview-book

深度学习面试宝典（含数学、机器学习、深度学习、计算机视觉、自然语言处理和SLAM等方向）

fb-ssem-dataset

The FB-SSEM dataset is a synthetic datasetconsisting of surround-view fisheye camera images and BEV maps from simulated sequences of ego car motion

flare_removal_pytorch

PyTorch Implementation of "How to Train Neural Networks for Flare Removal"

kuinameplates-wotlk

Kui_Nameplates for WoTLK.

lmffnet

Real-time semantic segmentation is widely used in the field of autonomous driving and robotics. Most previous networks achieved great accuracy based on a complicated model involving mass computing. The existing lightweight networks generally reduce the parameter sizes by sacrificing the segmentation accuracy. It is critical to balance the parameters and accuracy for real-time semantic segmentation tasks. In this paper, we introduce a Lightweight-Multiscale-Feature-Fusion Network (LMFFNet) mainly composed of three types of components: Split-Extract-Merge Bottleneck (SEM-B) block, Features Fusion Module (FFM), and Multiscale Attention Decoder (MAD). The SEM-B block extracts sufficient features with fewer parameters. FFMs fuse multiscale semantic features to effectively improve the segmentation accuracy. The MAD well recovers the details of the input images through the attention mechanism. Two networks combined with different components are proposed based on the LMFFNet model. Without pretraining, the smaller network of LMFFNet-S achieves 72.7% mIoU on Cityscapes test set at the 512×1024 resolution with only 1.1 M parameters at a reference speed of 98.9 fps running on a GTX1080Ti GPU while the larger version of LMFFNet-L achieves 74.7% mIoU with 1.4 M parameters at 89.6 fps. Besides, 67.7% mIoU at 208.9 fps and 70.3% mIoU at 72.4 fps are respectively achieved for 360 × 480 and 720 × 960 resolutions on CamVid test set using LMFFNet-S while LMFFNet--L achieves 68.1% mIoU at 182.9 fps and 71.0% mIoU at 66.5 fps, correspondingly. The proposed LMFFNets make an adequate trade-off between accuracy and parameter size for real-time inference for semantic segmentation tasks.

parking-slot-detection

shiq

Project for CVPR21 paper: "A Multi-Task Network for Joint Specular Highlight Detection and Removal".

tpvformer

An academic alternative to Tesla's occupancy network for autonomous driving.