项目类型

项目类别

最新

热搜企业:

Fork 0 关注 0

我们提供多种基于 OneFlow 的 NLP 模型和 NLP ops,包括: - TextCNN, Bi-LSTM for text classification - RoBERT

原始仓库地址:https://gitee.com/zhijiangtianshu/oneflow_nlp_model.git

浏览量:212 下载量:1 项目类别: 自然语言处理-模型
8 months前更新
Fork 0 关注 0

星际争霸的强化学习demo

原始仓库地址:https://gitee.com/zhijiangtianshu/oneflow_-alpha-star.git

浏览量:195 下载量:0 项目类别: 深度学习
8 months前更新
Fork 0 关注 0

OneFlow是开源的、采用全新架构设计,世界领先的工业级通用深度学习框架。 - 分布式训练全新体验,多机多卡如单机单卡一样简单 - 完美契合一站式平台(k8s + docker) - 原生支持超大模型 - 近零运行时开销、线性加速比 - 灵活支持多种深度学习编译器 - 自动混合精度 - 中立开放,合作面广 - 持续完善的算子集、模型库

原始仓库地址:https://gitee.com/zhijiangtianshu/oneflow.git

浏览量:235 下载量:0 项目类别: 深度学习
8 months前更新
Fork 0 关注 0

TS-VIS(天枢Vis)是[天枢人工智能开源开放平台](https://gitee.com/zhijiangtianshu/Dubhe)的可视化组件,支持Tensorflow、Pytorch、Oneflow等主流深度学习框架的可视化功能。 **[文档: https://feyily.github.io/tsvis-document/](https://feyily.github.io/tsvis-document/)** ![](docs/images/demo.gif) ## 亮点 * 框架无关,支持TensorFlow、PyTorch、OneFlow等主流深度学习框架可视化 * 超快的响应速度 * 支持大规模的可视化 * 支持训练过程实时可视化 * 支持降维分析样本可视化 * 支持神经网络异常可视化 ## 支持功能 - 模型结构可视化:可视化网络结构,包括计算图和结构图 - 标量数据可视化:可视化包括神经网络`accuary`和`loss`等的标量数据 - 媒体数据可视化:可视化包括图像、文字、音频在内的媒体数据 - 统计分析可视化:可视化神经网络中权重、偏置等的分布 - 降维分析可视化:通过降维算法,可视化任意高维数据 - 超参分析可视化:可视化不同超参数下的神经网络指标 - 异常检测可视化:将神经网络张量数据映射到二维,可视化张量数据统计信息 - 用户定制可视化:可以将所有功能移动到该模块进行可视化

原始仓库地址:https://gitee.com/zhijiangtianshu/zjvis.git

浏览量:213 下载量:3 项目类别: 深度学习
8 months前更新
to-fu Python
Fork 0 关注 3

Toolbox for Few-Shot Learning

原始仓库地址:https://toscode.gitee.com/pkumlg/to-fu.git

浏览量:302 下载量:0 项目类别: 计算机视觉
11 months前更新
open-exchange Python
Fork 0 关注 1

open-exchange is a model conversion and visualization tool to help users inter-operate among different deep learning frameworks. Convert models between PyTorch and Tensorflow.

原始仓库地址:https://github.com/icgy96/open-exchange.git

浏览量:298 下载量:0 项目类别: 深度学习
11 months前更新
duo N/A
Fork 0 关注 0

# Detecting Underwater Objects (DUO) 水下目标检测技术引起了人们的越来越多的关注。然而,这个领域仍然存在着若干挑战。我们通过应对以下挑战,来促进水下目标检测技术的发展。首先,由于目前可用的数据集基本上缺乏测试集的真值标注文件,这导致研究者必须在训练集上划分出测试集来与其他方法进行比较。训练其他方法会增加工作量;不同的研究人员会划分不同的测试集,导致这一领域没有一个统一的基准来比较不同算法的性能。其次,这些已有数据集(URPC系列)也存在其他缺点,如相似图像过多或标签标注不准确。针对以上这些挑战,我们在对所有相关数据集进行收集和重新标注的基础上,引入了一个新的数据集——水下目标检测数据集(Detection Underwater Objects, DUO)和其相应的基准(benchmark)。DUO包含了多种多样的水下图像,并且具有更合理的注释。相应的基准为学术研究和工业应用提供了多种目标检测模型(在mmddetection框架下)在DUO上的效率和准确性等指标对比数据,其中NVIDIA嵌入式平台JETSON AGX XAVIER也被用于评估不同检测模型的实时推理速度,用以模拟机器人的嵌入式环境。 ## 下载 * [BaiduYun](https://pan.baidu.com/s/1Be8zc9UdR_Pdsyotg_vR2Q) Key : 4bfl ## 引用 @ARTICLE{2021arXiv210605681L, author = {{Liu}, Chongwei and {Li}, Haojie and {Wang}, Shuchang and {Zhu}, Ming and {Wang}, Dong and {Fan}, Xin and {Wang}, Zhihui}, title = "{A Dataset And Benchmark Of Underwater Object Detection For Robot Picking}", journal = {arXiv e-prints}, year = 2021, month = jun, eid = {arXiv:2106.05681}, pages = {arXiv:2106.05681}, archivePrefix = {arXiv}, eprint = {2106.05681}, primaryClass = {cs.CV} }

原始仓库地址:https://github.com/chongweiliu/duo.git

浏览量:339 下载量:0 项目类别: 图像
11 months前更新
Fork 0 关注 0

# UDD_OFFICIAL 为促进海洋养殖场水下自动抓取机器人的发展,大连理工大学多媒体实验室(MM_lab)提出了一个水下海洋牧场目标检测数据集UDD。UDD包括3个类别(海参、海胆和扇贝)共2,227张图片。这是第一个在真实海洋牧场中采集到的数据集。我们还提出了一个新的泊松混合生成对抗网络(Poisson GAN)。 利用Poisson GAN我们构造了一个大型包含18,000张图像的增强数据集(AUDD)。此外,为了使检测器更好的适应水下抓取环境,我们还合成了一个预训练数据集(pre-training dataset),包含59万张图像。 ## Download * [BaiduYun](https://pan.baidu.com/s/1byq7wEID-OzLSJ8p5A6Z5g) Key : 2kse ## Citation @article{Wang:2020ug, author = {Wang, Zhihui and Liu, Chongwei and Wang, Shijie and Tang, Tao and Tao, Yulong and Yang, Caifei and Li, Haojie and Liu, Xing and Fan, Xin}, title = {{UDD: An Underwater Open-sea Farm Object Detection Dataset for Underwater Robot Picking}}, journal = {arXiv.org}, year = {2020}, eprint = {2003.01446v1}, eprinttype = {arxiv}, eprintclass = {cs.CV}, month = mar, annote = {10 pages, 9 figures} }

原始仓库地址:https://github.com/chongweiliu/udd_official.git

浏览量:500 下载量:0 项目类别: 图像
about 1 year前更新

**Data Description** The detection research of sonar images has a wide range of application value in many fields such as industry, environment and military. Sonar image data is often acquired together with water depth detection and bottom detection data, which enables us to observe the shallow structure of the seabed. This data set was launched by Pengcheng Laboratory, which is currently the largest and most extensive acoustic image data set in the industry. Please contact with yangj01@pcl.ac.cn if you have any questions. ** Download** *Baidu Cloud* link: https://pan.baidu.com/s/1X0Pcl1E-ctRkgJDpQowZEw password: 5lq2 *OneDrive* https://1drv.ms/u/s!AlpdPhejQ4FqiDL25YVhjxxT1WeU?e=DcwBRD **Training Set** The training set contains 4000 forward-looking sonar image files in .bmp format and corresponding annotation result files in .xml format. The file organization structure is as follows: ![](/attachments/download/968) The image data in the forward-looking sonar image file represents the acoustic reflection intensity information, and the principle is as follows: ![](/attachments/download/969) The pixel with the coordinate ![](/attachments/download/978) in the image represents the acoustic reflection intensity information at the direction ![](/attachments/download/976) and distance ![](/attachments/download/977) in the polar coordinate system. Among them, ![](/attachments/download/979) and ![](/attachments/download/980) respectively represent the horizontal angle and slant range of the forward looking sonar, ![](/attachments/download/981) and ![](/attachments/download/982) represent the horizontal and vertical size of the image respectively. The annotation result file contains the detection frame parameters (position and size) and target type of the target in the corresponding sonar image. Target types include cube, ball, cylinder, human body, tyre, circle cage, square cage, and Metal bucket and other 8 categories, the label format is as follows: ![](/attachments/download/970) The </sonar> field contains the working parameters of the sonar when the corresponding image is collected, such as the horizontal/vertical angle of the sonar, the range of the sonar slope, and the working frequency of the sonar. **Test Set** Test Set includes A_Board and B_Board, which contains 1000 sonar image in .bmp format and corresponding sonar annotation files in .xml respectively. The sonar parameter information file format is as follows: ![](/attachments/download/971) **Extra Information:** The data was collected using Tritech Gemini 1200i forward looking sonar. This data set is the original echo intensity information of the sonar in the form of a two-dimensional matrix, which is stored as a bmp image format for the convenience of process and can be read as 3-channel image data. The data of the 3 channels are equal, so the data of one channel can be used in actual processing; The original sonar image has not been adjusted for gain and is directly displayed as a nearly black image. It is suggested that the display effect of the data after histogram equalization is better, which helps to establish an intuitive understanding of the sonar image, but whether to add gain to the data should not have much influence on the algorithm design (the histogram equalization algorithm can be moved Discussion area for reference) **Data Example** ![](/attachments/download/972) The picture above is a forward-looking sonar image. The red box is the target marking frame. From left to right, they are square cage, ball, and tyre. ![](/attachments/download/973) The picture above is a forward-looking sonar image. The red box is the target marking frame. From left to right, they are human body, ball, and circle cage. ![](/attachments/download/974) The picture above is a forward-looking sonar image. The red box is the target marking frame. From left to right, they are tyre and metal bucket.

浏览量:1571 下载量:0 项目类别: 图像
about 1 year前更新
xiangshan Scala
Fork 0 关注 0

Open-source high-performance RISC-V processor

原始仓库地址:https://github.com/openxiangshan/xiangshan.git

浏览量:570 下载量:5 项目类别: RISC-V
about 1 year前更新

ihub@pcl.ac.cn 鹏城实验室人工智能研究中心

版权所有:鹏城实验室    粤ICP备18066427号-6    Powerd by 国防科技大学Trustie