数据集

BDD-100K - Berkeley DeepDrive 数据集，包含 100,000 个视频序列的驾驶场景数据

自动驾驶数据集

https://huggingface.co/datasets/drive-bench/arena/tree/main

https://github.com/OpenDriveLab/DriveLM/blob/main/docs/data_prep_nus.md

Tiu MoLess than 1 minute

Awesome Hyperbolic Representation Learning

A curated list of resources dedicated to hyperbolic representation learning and its applications across various domains. Hyperbolic spaces are particularly suited for representing hierarchical data and tree-like structures, offering advantages over Euclidean embeddings for many complex data types.

Tiu MoAbout 9 min

"""
@author yutanglee
@description 这个代码可以免浏览器拨号上网，原理很简单，就是给认证服务器发送一个http请求。
@use 你需要提供上网账号、密码和运营商，是否采用手机端或者PC端是可选的。手机端和PC端是指，重邮的每个上网账号可以同时存在一个PC账号和手机端。
"""
from ast import arg
from pydoc import describe
import requests
import subprocess
import re
import argparse
import base64


def login(usr, psw, isp, ip, mobile=True):
    
    url = 'http://192.168.200.2:801/eportal/?c=Portal&a=login&callback=dr1003&login_method=1&' +\
        'user_account=%2C0%2C' + usr + '%40' + isp +'&user_password=' + psw + '&wlan_user_ip=' + ip + \
        '&wlan_user_ipv6=&wlan_user_mac=000000000000&wlan_ac_ip=&wlan_ac_name=&jsVersion=3.3.3&v=10390'
    if not mobile:
        headers = {
            'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/114.0.0.0 Safari/537.36 Edg/114.0.1823.58'}
    else:
        headers = {
            'User-Agent': 'Mozilla/5.0 (Linux; Android 6.0; Nexus 5 Build/MRA58N) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/114.0.0.0 Mobile Safari/537.36 Edg/114.0.1823.58'}
    r = requests.get(url=url, headers=headers)
    return r.text

def get_local_ip():
    try:
        # 运行 ip a 命令获取网络接口信息
        output = subprocess.check_output(["ip", "a"]).decode("utf-8")
        
        # 使用正则表达式解析 IP 地址
        ip_pattern = r"inet (\d+\.\d+\.\d+\.\d+)"
        match = re.findall(ip_pattern, output)
        
        # 筛选出以特定 IP 段开头的 IP 地址
        target_ips = [ip for ip in match if ip.startswith("10.16.") or ip.startswith("10.20.")]
        
        if target_ips:
            return target_ips[0]  # 返回第一个满足条件的 IP 地址
        else:
            return None
    except subprocess.CalledProcessError:
        return None

    
if __name__ == '__main__':
    args = argparse.ArgumentParser()
    args.add_argument('--user', type=str, default='<your_id>', help='your account for CQUPT network, maybe is your id number login ehall.cqupt.edu.cn')
    args.add_argument('--psw', type=str, default='<your_password>', help='password')
    args.add_argument('--isp', type=str, default='<your_isp>', help='cmcc|telecom|unicom|xyw')
    args.add_argument('--mobile', action='store_true', help='do you want use mobile mode?')
    args = args.parse_args()
    
    if args.isp != 'cmcc' and args.isp != 'telecom' and args.isp != 'unicom' and args.isp != 'xyw':
        print('运营商只支持cmcc|telecom|unicom|xyw  重新检查输入！')
        exit(1)

    ip = get_local_ip()
    if ip is None:
        ip = input("自动获取IP失败, 请你输入正确的IP: ")
    print('your ip is:', ip)
    print('your user is:', args.user)
    print('your password is:', args.psw)    
    print('your isp is:', args.isp)
    args.mobile = True

    response = login(args.user, args.psw, args.isp, ip, args.mobile)
    start_index = response.find('"result":"') + len('"result":"')
    end_index = response.find('"', start_index)
    result_value = response[start_index:end_index]
    if result_value == '1':
        print('登录成功！')
    else:
        start_index = response.find('"msg":"') + len('"msg":"')
        end_index = response.find('"', start_index)
        msg_encoded = response[start_index:end_index]
        # 对msg进行Base64解码
        msg_decoded = base64.b64decode(msg_encoded).decode('utf-8')

        print(msg_decoded, response)
        if msg_decoded == 'ldap auth error':
            print('密码错误！')
        elif msg_decoded == 'userid error1':
            print('账号不存在！')

Tiu MoAbout 2 min

1、题目： MSC-Bench: Benchmarking and Analyzing Multi-Sensor Corruption for Driving Perception

链接： https://t.zsxq.com/BRSA3

简介： MSC-Bench: 第一个针对多传感器自动驾驶感知模型在各种传感器损坏情况下的鲁棒性进行评估的综合基准

时间： 2025-01-10T23:52:48.526+0800

2、题目： Hidden Biases of End-to-End Driving Datasets

链接：https://t.zsxq.com/BRSA3

Tiu MoAbout 4 min

Hyperbolic Space Learning Resources

双曲空间学习资源

A comprehensive collection of resources for hyperbolic space learning, representation, and deep learning implementations. 双曲空间学习、表示和深度学习实现的综合资源集合。

Computer Vision & Image Processing

计算机视觉与图像处理

HyperbolicCV - Hyperbolic Computer Vision implementation
- 双曲空间中的计算机视觉实现，用于处理视觉数据
Hyperbolic Image Embeddings - Image embeddings in hyperbolic space
- 在双曲空间中生成图像嵌入表示

Tiu MoAbout 1 min

https://github.com/Thinklab-SJTU/Awesome-LLM4AD 综述地址

Trajectory-guided Control Prediction for End-to-end Autonomous Driving: A Simple yet Strong Baseline. NeurIPS 22 Think Twice before Driving: Towards Scalable Decoders for End-to-End Autonomous Driving. CVPR 23 DriveAdapter: Breaking the Coupling Barrier of Perception and Planning in End-to-End Autonomous Driving. ICCV 23 Oral Think2drive: Efficient reinforcement learning by thinking in latent world model for quasi-realistic autonomous driving (in carla-v2). ECCV 24

Tiu MoLess than 1 minute

R1 Zero GRPO Resources

A curated collection of resources related to R1 Zero and GRPO (Generative Reward-Penalty Optimization) implementations and research.