Quadrupedal robots are conquering various indoor and outdoor applications due to their ability to navigate challenging uneven terrains. Exteroceptive information greatly enhances this capability since perceiving their surroundings allows them to adapt their controller and thus achieve higher levels of robustness. However, sensors such as LiDARs and RGB cameras do not provide sufficient information to quickly and precisely react in a highly dynamic environment since they suffer from a bandwidth-latency tradeoff. They require significant bandwidth at high frame rates while featuring significant perceptual latency at lower frame rates, thereby limiting their versatility on resource-constrained platforms. In this work, we tackle this problem by equipping our quadruped with an event camera, which does not suffer from this tradeoff due to its asynchronous and sparse operation. In leveraging the low latency of the events, we push the limits of quadruped agility and demonstrate high-speed ball catching for the first time. We show that our quadruped equipped with an event camera can catch objects with speeds up to 15 m/s from 4 meters, with a success rate of 83%. Using a VGA event camera, our method runs at 100 Hz on an NVIDIA Jetson Orin.
相關內容
Wide deployment of deep neural networks (DNNs) based applications (e.g., style transfer, cartoonish), stimulating the requirement of copyright protection of such application's production. Although some traditional visible copyright techniques are available, they would introduce undesired traces and result in a poor user experience. In this paper, we propose a novel plug-and-play invisible copyright protection method based on defensive perturbation for DNN-based applications (i.e., style transfer). Rather than apply the perturbation to attack the DNNs model, we explore the potential utilization of perturbation in copyright protection. Specifically, we project the copyright information to the defensive perturbation with the designed copyright encoder, which is added to the image to be protected. Then, we extract the copyright information from the encoded copyrighted image with the devised copyright decoder. Furthermore, we use a robustness module to strengthen the decoding capability of the decoder toward images with various distortions (e.g., JPEG compression), which may be occurred when the user posts the image on social media. To ensure the image quality of encoded images and decoded copyright images, a loss function was elaborately devised. Objective and subjective experiment results demonstrate the effectiveness of the proposed method. We have also conducted physical world tests on social media (i.e., Wechat and Twitter) by posting encoded copyright images. The results show that the copyright information in the encoded image saved from social media can still be correctly extracted.
Serverless computing is a popular cloud computing paradigm, which requires low response latency to handle on-demand user requests. There are two prominent techniques employed for reducing the response latency: keep fully initialized containers alive (Warm Container) or reduce the new container startup (cold start) latency. This paper presents the 3rd container startup mode: Hibernate Container, which starts faster than the cold start container mode and consumes less memory than the Warm Container mode. Hibernate Container is essentially a "deflated" Warm Container. Its application memory is swapped out to disk, the freed memory is reclaimed and file based mmap memory is cleaned-up. The Hibernate Container's deflated memory is inflated in response to user requests. As Hibernate Container's application is fully initialized, its response latency is less than the cold start mode; and as the application memory is deflated, its memory consumption is less than the Warm Container mode. Additionally, when a Hibernate Container is "woken up" to process a request, the Woken-up Container has similar response latency to Warm Container but less memory consumption because not all the deflated memory needs to be inflated. We implemented the Hibernate technique as part of the open source Quark secure container runtime project and our test demonstrated that Hibernate Container consumes about 7\% to 25\% of the Warm Container memory. All of this results in a higher deployment density, lower latency and appreciable improvements in the overall system performance.
隨(sui)著無(wu)(wu)(wu)人(ren)駕駛(shi)飛(fei)行器(qi)（UAVs），也被(bei)稱為(wei)無(wu)(wu)(wu)人(ren)機(ji)，變得(de)(de)容易(yi)獲得(de)(de)和負(fu)擔得(de)(de)起，這些(xie)設備的(de)(de)應(ying)用(yong)(yong)已經大(da)大(da)增加(jia)。其中一種(zhong)應(ying)用(yong)(yong)是使用(yong)(yong)無(wu)(wu)(wu)人(ren)機(ji)飛(fei)越大(da)面積(ji)區域并(bing)探測(ce)所(suo)需實體。例如，一群(qun)(qun)無(wu)(wu)(wu)人(ren)機(ji)可以(yi)(yi)(yi)探測(ce)海洋表面附近的(de)(de)海洋生(sheng)物(wu)，并(bing)向用(yong)(yong)戶提供發現的(de)(de)動物(wu)的(de)(de)位置(zhi)和類型。然(ran)而，即使無(wu)(wu)(wu)人(ren)機(ji)技(ji)術的(de)(de)成(cheng)(cheng)本降低，由于使用(yong)(yong)內置(zhi)先進功能的(de)(de)定制硬件，這種(zhong)應(ying)用(yong)(yong)的(de)(de)成(cheng)(cheng)本也很高。因此(ci)，本論文的(de)(de)重點(dian)是編制一個容易(yi)定制的(de)(de)、低成(cheng)(cheng)本的(de)(de)無(wu)(wu)(wu)人(ren)機(ji)設計，并(bing)配備必要(yao)(yao)的(de)(de)硬件，以(yi)(yi)(yi)實現自主行為(wei)、蜂群(qun)(qun)協調和機(ji)載物(wu)體探測(ce)能力。此(ci)外，本論文概述了必要(yao)(yao)的(de)(de)網絡結構，以(yi)(yi)(yi)處理無(wu)(wu)(wu)人(ren)機(ji)群(qun)(qun)的(de)(de)互連和帶寬要(yao)(yao)求。
無人(ren)機機載系統使(shi)(shi)用(yong)PixHawk 4飛(fei)行(xing)控制器來處理飛(fei)行(xing)機械，使(shi)(shi)用(yong)Raspberry Pi 4作為(wei)通(tong)(tong)(tong)用(yong)計算能力的(de)配套(tao)計算機，并(bing)使(shi)(shi)用(yong)NVIDIA Jetson Nano開發套(tao)件來實(shi)時進行(xing)物體檢測。實(shi)施的(de)網(wang)(wang)絡(luo)(luo)遵(zun)循(xun)802.11s標準，采(cai)用(yong)HWMP路由協議進行(xing)多(duo)跳通(tong)(tong)(tong)信。這種拓撲結(jie)構允許無人(ren)機通(tong)(tong)(tong)過網(wang)(wang)絡(luo)(luo)轉發數(shu)據包，大(da)大(da)擴展了蜂群的(de)飛(fei)行(xing)范(fan)圍。我們的(de)實(shi)驗表明，所選的(de)硬件和實(shi)現的(de)網(wang)(wang)絡(luo)(luo)可(ke)以(yi)在高達(da)1000英(ying)尺的(de)范(fan)圍內提(ti)供直接(jie)的(de)點對點通(tong)(tong)(tong)信，通(tong)(tong)(tong)過信息轉發可(ke)以(yi)擴大(da)范(fan)圍。該(gai)網(wang)(wang)絡(luo)(luo)還為(wei)帶寬密集(ji)型(xing)數(shu)據（如實(shi)時視頻流）提(ti)供了足(zu)夠的(de)帶寬。預(yu)計飛(fei)行(xing)時間約(yue)為(wei)17分(fen)鐘，擬(ni)議的(de)設計為(wei)中程空中監視應用(yong)提(ti)供了低成(cheng)本的(de)無人(ren)機群解決方案。
付費5元查看完整內容
Recently, ChatGPT, along with DALL-E-2 and Codex,has been gaining significant attention from society. As a result, many individuals have become interested in related resources and are seeking to uncover the background and secrets behind its impressive performance. In fact, ChatGPT and other Generative AI (GAI) techniques belong to the category of Artificial Intelligence Generated Content (AIGC), which involves the creation of digital content, such as images, music, and natural language, through AI models. The goal of AIGC is to make the content creation process more efficient and accessible, allowing for the production of high-quality content at a faster pace. AIGC is achieved by extracting and understanding intent information from instructions provided by human, and generating the content according to its knowledge and the intent information. In recent years, large-scale models have become increasingly important in AIGC as they provide better intent extraction and thus, improved generation results. With the growth of data and the size of the models, the distribution that the model can learn becomes more comprehensive and closer to reality, leading to more realistic and high-quality content generation. This survey provides a comprehensive review on the history of generative models, and basic components, recent advances in AIGC from unimodal interaction and multimodal interaction. From the perspective of unimodality, we introduce the generation tasks and relative models of text and image. From the perspective of multimodality, we introduce the cross-application between the modalities mentioned above. Finally, we discuss the existing open problems and future challenges in AIGC.
隨著大規模作戰(zhan)(zhan)行(xing)動(dong)（LSCO）的(de)決定性(xing)作戰(zhan)(zhan)訓練環(huan)境(jing)（DATE）場景變(bian)(bian)得(de)(de)更(geng)加(jia)(jia)復(fu)雜，敵對(dui)(dui)勢力（OPFOR）繼(ji)續(xu)變(bian)(bian)得(de)(de)更(geng)加(jia)(jia)適應獵(lie)(lie)殺和(he)(he)瞄(miao)準(zhun)(zhun)藍軍(jun)(jun)指揮所和(he)(he)關鍵資(zi)產(chan)，如整(zheng)個(ge)戰(zhan)(zhan)斗(dou)空(kong)間的(de)炮(pao)(pao)(pao)兵設備和(he)(he)反(fan)火力（CF）雷達(da)。理解(jie)這一點(dian)至關重要，因(yin)為(wei)雖然美軍(jun)(jun)高度重視使(shi)(shi)用(yong)無(wu)人(ren)機系統(tong)(UAS)進行(xing)信(xin)息、監視和(he)(he)偵察(ISR)，以(yi)確定敵人(ren)在(zai)時(shi)(shi)間和(he)(he)空(kong)間上可(ke)能有高收益目(mu)標(HPTs)的(de)位(wei)置，但(dan)OPFOR可(ke)以(yi)利用(yong)這一點(dian)并取(qu)得(de)(de)成功，因(yin)為(wei)一個(ge)簡單的(de)事實:美軍(jun)(jun)使(shi)(shi)用(yong)紅(hong)方UAS進行(xing)狩獵(lie)(lie)，而且(qie)他(ta)們(men)(men)(men)擅長于此。OPFOR不(bu)需要像(xiang)美軍(jun)(jun)那樣使(shi)(shi)用(yong)紅(hong)方UAS進行(xing)大量的(de)信(xin)息收集（IC），因(yin)為(wei)OPFOR明白，藍軍(jun)(jun)的(de)火炮(pao)(pao)(pao)和(he)(he)雷達(da)在(zai)大范圍內的(de)移動(dong)足跡不(bu)多，因(yin)此，一旦他(ta)們(men)(men)(men)找到(dao)HPT，他(ta)們(men)(men)(men)可(ke)以(yi)很容易地使(shi)(shi)用(yong)紅(hong)方UAS與特種部(bu)隊（SPF）配合，用(yong)遠程精確火炮(pao)(pao)(pao)獵(lie)(lie)殺、瞄(miao)準(zhun)(zhun)和(he)(he)攻(gong)擊藍軍(jun)(jun)的(de)關鍵資(zi)產(chan)。如果師(shi)炮(pao)(pao)(pao)兵（DIVARTY）、軍(jun)(jun)團野戰(zhan)(zhan)炮(pao)(pao)(pao)兵旅（FAB）和(he)(he)旅級直接(jie)支援(yuan)（DS）營(ying)都有自(zi)己的(de)目(mu)標定位(wei)UAS分(fen)(fen)隊，那會怎樣？這將縮短從傳感器到(dao)射手(shou)的(de)殺傷鏈，減少目(mu)標逃脫概率，減少目(mu)標反(fan)應時(shi)(shi)間，減少動(dong)態重新分(fen)(fen)配給各(ge)師(shi)和(he)(he)軍(jun)(jun)團情報優先事項的(de)ISR資(zi)產(chan)需要，提高獵(lie)(lie)殺、瞄(miao)準(zhun)(zhun)和(he)(he)塑造敵人(ren)炮(pao)(pao)(pao)兵縱深的(de)有效性(xing)，同時(shi)(shi)不(bu)干擾他(ta)們(men)(men)(men)各(ge)自(zi)S2/G2參謀部(bu)的(de)IC工作。美國陸軍(jun)(jun)在(zai)如何分(fen)(fen)配無(wu)人(ren)機方面的(de)這一革命性(xing)和(he)(he)根本性(xing)的(de)轉(zhuan)變(bian)(bian)能否為(wei)擁有和(he)(he)主(zhu)導狩獵(lie)(lie)提供解(jie)決方案，以(yi)對(dui)(dui)抗(kang)一個(ge)近乎對(dui)(dui)等的(de)OPFOR對(dui)(dui)手(shou)，后者相信(xin)通(tong)過火炮(pao)(pao)(pao)和(he)(he)綜(zong)合防空(kong)可(ke)戰(zhan)(zhan)勝對(dui)(dui)手(shou)贏得(de)(de)戰(zhan)(zhan)斗(dou)?
本文將(jiang)討(tao)論(lun)一個理論(lun)，即(ji)軍團的(de)(de)(de)FAB、DIVARTY和(he)BCT DS野戰(zhan)炮(pao)(pao)(pao)兵(bing)(FA)BN，如(ru)果(guo)獲(huo)得了由一個灰鷹(GE)排(pai)組(zu)成的(de)(de)(de)目(mu)標定位(wei)(wei)分(fen)隊的(de)(de)(de)作(zuo)戰(zhan)控(kong)制(zhi)權(quan)(OPCON) 。以(yi)(yi)及必要的(de)(de)(de)人(ren)員(yuan)來(lai)進行(xing)開發，允(yun)許炮(pao)(pao)(pao)兵(bing)部隊指(zhi)揮(hui)官(guan)擁(yong)有(you)師和(he)兵(bing)團指(zhi)揮(hui)官(guan)的(de)(de)(de)目(mu)標定位(wei)(wei)過程(cheng)。這(zhe)個解(jie)決方案可(ke)以(yi)(yi)確保(bao)野戰(zhan)炮(pao)(pao)(pao)兵(bing)部隊能夠(gou)打擊目(mu)標，削弱(ruo)敵人(ren)的(de)(de)(de)遠程(cheng)火炮(pao)(pao)(pao)，瓦(wa)解(jie)綜合防空(kong)能力，并提高殺傷鏈(lian)的(de)(de)(de)有(you)效性，以(yi)(yi)滿足其指(zhi)揮(hui)官(guan)的(de)(de)(de)作(zuo)戰(zhan)重點。本文還(huan)將(jiang)從條(tiao)令、組(zu)織、訓練(lian)、材(cai)料(liao)、領導和(he)教育、人(ren)員(yuan)、設(she)施(shi)和(he)政策(ce)（DOTMLPF-P）的(de)(de)(de)模式來(lai)看(kan)待這(zhe)個問題，以(yi)(yi)提供一個整體的(de)(de)(de)視(shi)角(jiao)來(lai)看(kan)待從可(ke)能的(de)(de)(de)訓練(lian)概念到陸軍范(fan)圍內的(de)(de)(de)實施(shi)建議(yi)。
付費5元查看完整內容
題目： Instance-aware, Context-focused, and Memory-efficient Weakly Supervised Object Detection
摘要：
弱監(jian)(jian)督(du)學習通過減少訓(xun)練(lian)過程中(zhong)對(dui)(dui)強監(jian)(jian)督(du)的(de)(de)(de)需(xu)求，已經(jing)成為一(yi)(yi)種引人注目的(de)(de)(de)對(dui)(dui)象(xiang)檢(jian)測工(gong)具。然而，主要的(de)(de)(de)挑戰仍然存在:(1)對(dui)(dui)象(xiang)實例的(de)(de)(de)區(qu)分(fen)可能是(shi)(shi)模糊的(de)(de)(de);(2)探測器(qi)往往聚焦(jiao)于有區(qu)別(bie)的(de)(de)(de)部分(fen)，而不是(shi)(shi)整個(ge)物體;(3)如果(guo)準確(que)性不高，對(dui)(dui)象(xiang)建議對(dui)(dui)于高回憶來說是(shi)(shi)冗(rong)余(yu)的(de)(de)(de)，這會導致(zhi)大量(liang)的(de)(de)(de)內存消(xiao)耗。解(jie)決這些(xie)挑戰是(shi)(shi)困難的(de)(de)(de)，因為它經(jing)常需(xu)要消(xiao)除不確(que)定性和(he)瑣碎的(de)(de)(de)解(jie)決方案。為了解(jie)決這些(xie)問題，我(wo)們(men)開發了一(yi)(yi)個(ge)實例感知和(he)上下文(wen)相關的(de)(de)(de)統(tong)一(yi)(yi)框架(jia)。它采用了一(yi)(yi)個(ge)實例感知的(de)(de)(de)自(zi)訓(xun)練(lian)算(suan)法和(he)一(yi)(yi)個(ge)可學習的(de)(de)(de)具體DropBlock，同時設(she)計(ji)了一(yi)(yi)個(ge)內存有效的(de)(de)(de)順序批(pi)處理反(fan)向(xiang)傳播。我(wo)們(men)提出的(de)(de)(de)方法在COCO(12.1%的(de)(de)(de)AP, 24.8%的(de)(de)(de)AP50)、VOC 2007(54.9%的(de)(de)(de)AP)和(he)VOC 2012(52.1%的(de)(de)(de)AP)上取得了最先進的(de)(de)(de)結果(guo)，極大地改善了基(ji)線(xian)。此外，該(gai)方法是(shi)(shi)第一(yi)(yi)個(ge)對(dui)(dui)基(ji)于ResNet的(de)(de)(de)模型和(he)弱監(jian)(jian)督(du)視頻(pin)對(dui)(dui)象(xiang)檢(jian)測進行基(ji)準測試的(de)(de)(de)方法。
付費5元查看完整內容
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191