TOP COLUMN 「CVPR2025-速報-」ResearchPortトップカンファレンス定点観測 vol.17

「CVPR2025-速報-」ResearchPortトップカンファレンス定点観測 vol.17

computer vision CVPR トップカンファレンス定点観測
2025年6月9日 10時00分 公開

本記事3行要約：

● 今年も投稿数(13,008)・採択数(2,872)ともに過去最多更新
● 日本人研究者からの発表件数・著者数も昨年超え(74件/のべ134名)
● 今年もdiffusion・language modelが多数出現、Gaussian splattingも躍進


トップカンファレンス定点観測シリーズ vol.17、「CVPR」の第5弾です。

これまでResearchPortでは、コンピュータビジョン分野で最高峰の会議であるCVPR（IEEE/CVF Conference on Computer Vision and Pattern Recognition）における論文投稿数の増加とそれに伴う盛況感について2021・2022・2023年(速報)版・2024年(速報)版を記事にしてきました。

＊参照：
「CVPR」ResearchPortトップカンファレンス定点観測 vol.1
「CVPR2022」ResearchPortトップカンファレンス定点観測 vol.3
「CVPR2023-速報-」ResearchPortトップカンファレンス定点観測 vol.9
「CVPR2024-速報-」ResearchPortトップカンファレンス定点観測 vol.13

今年も6月となり、いよいよCVPR2025が開催されます。
最多論文数更新は今年度も変わらず、日本からの発表も伸びております。本会議が開催される直前に、速報記事*1として統計情報をまとめました。

CVPR2025　開催概要
▶ 開催期間：　11-15 Jun., 2025
▶ 開催都市：　Nashville TN, USA
▶ 公式HP：　 https://cvpr.thecvf.com/Conferences/2025

＊1　2025年6月4日時点

■CVPR2025総括

前年同様、CVPR2025 でも投稿数・採択数ともに過去最高数を更新しております。投稿数は13,000件を超え、相変わらずの盛況感を反映している数値となっています。
詳細は以前の記事を参照していただきたく、早速本題に入りたいと思います。CVPR2025最新数値は以下に示しております（表1・図1）。

Year #papers #orals/
highlights		 #submissions acceptance rate oral/highlights
acceptance rate		 Venue
1996 137 73 551 24.86% 13.25% San Francisco,CA
1997 173 62 544 31.80% 11.40% San juan,Puerto Rico
1998 139 42 453 30.68% 9.27% Santa Barbara,CA
1999 192 73 503 38.17% 14.51% Fort Collins,CO
2000 220 66 466 47.21% 14.16% Hilton Head,SC
2001 273 78 920 29.67% 8.48% Kauai,HI
2002 – – – – – –
2003 209 60 905 23.09% 6.63% Madison,WI
2004 260 54 873 29.78% 6.19% Washington,DC
2005 326 75 1,160 28.10% 6.47% San Diego,CA
2006 318 54 1,131 28.12% 4.77% New York,NY
2007 353 60 1,250 28.24% 4.80% Minneapolis,MN
2008 508 64 1,593 31.89% 4.02% Anchorage,AK
2009 384 61 1,464 26.23% 4.17% Miami,FL
2010 462 78 1,724 26.80% 4.52% San Francisco,CA
2011 436 59 1,677 26.00% 3.52% Colorado Springs,CO
2012 463 48 1,933 23.95% 2.48% Providence
2013 471 60 1,798 26.20% 3.34% Portland,OR
2014 540 104 1,807 29.88% 5.76% Columbus,OH
2015 602 71 2,123 28.36% 3.34% Boston,MA
2016 643 83 2,145 29.98% 3.87% Las Vegas,NV
2017 783 71 2,680 29.22% 2.65% Hawaii,HW
2018 979 70 3,359 29.15% 2.08% Salt Lake City,UT
2019 1,294 288 5,160 25.08% 5.58% Long Beach,CA
2020 1,466 335 5,865 25.00% 5.71% Seattle,WA
2021 1,660 295 7,015 23.66% 4.21% Nashville,TN
2022 2,063 342 8,161 25.28% 4.19% New Orleans, LA
2023 2,357 247 9,155 25.75% 2.70% Vancouver, Canada
2024 2,716 324 11,532 23.55% 2.81% Seattle,WA
2025 2,872 335 13,008 22.08% 2.58% Nashville, TN

表1　CVPR論文投稿数および採択率

＊ 2023年のoralはhighlightsとaward candidateを合わせた件数です。
＊ 出典：https://cvpr.thecvf.com/Conferences/2025/AcceptedPapers
　

図1　CVPR統計推移

今年は、昨年比+1,500件程増加の13,008件の論文が投稿され、うち2,872件が採択されました。このうち335件が口頭発表論文（oral/highlights）です。つまり採択率は22.08%で、oral/highlights採択率は2.58%となります。

［CVPR2025　統計］
　・論文投稿数：　13,008件
　・採択数：　2,872件（oral/highlights採択 335件）
　・採択率：　23.55%
　・oral/highlights採択率：　2.58%

■日本人研究者の活躍

今年度も、日本人著者の比率も調べておりますので、結果を以下に示します（表2）。

全著者数 17,804名に占める日本人の割合は、2025年において0.75%となっております。
日本人研究者の比率を2024年と比べると、著者数 130→134 、日本人含む論文数 66→74 と増加しているものの、全体の採択論文・平均著者数ともに大きく増加しているため、著者比率としては0.1%下落という結果になっております。
日本人著者数は今年も増加となっており、今年も変わらず国内コンピュータビジョン研究の盛り上がりを感じられる数字ではないでしょうか。

開催年 採択論文数 著者数 平均著者数 日本人
著者数		 日本人
著者比率		 日本人著者
を含む論文		 日本人著者が
絡む論文比率
2014 540 1,881 3.48 45 2.39% 26 4.81%
2015 602 2,207 3.67 33 1.50% 20 3.32%
2016 643 2,487 3.87 45 1.81% 21 3.27%
2017 783 3,185 4.07 61 1.92% 29 3.70%
2018 979 4,214 4.30 93 2.21% 38 3.88%
2019 1,294 5,863 4.53 86 1.47% 40 3.09%
2020 1,466 6,970 4.75 65 0.93% 38 2.59%
2021 1,660 8,087 4.87 72 0.89% 42 2.53%
2022 2,063 10,874 5.27 108 0.99% 52 2.52%
2023 2,357 12,722 5.40 117 0.92% 60 2.55%
2024 2,716 15,288 5.63 130 0.85% 66 2.43%
2025 2,872 17,804 6.20 134 0.75% 74 2.58%

表2　CVPR投稿論文全体の著者数に占める日本人比率の推移

CVPR2025 個人別採択件数
著者 採択数
Shunsuke Saito
6
Masayoshi Tomizuka
4
Yuki Mitsufuji
3
Imari Sato
3
Ryo Hachiuma
2
Tatsuya Harada
2
Daisuke Iso
2
Norimasa Kobori
2
Ryota Maeda
2
Koki Nagano
2
Ko Nishino
2
Takashi Shibuya
2
Seiichi Uchida
2
Hiroki Adachi
1
Kazuki Adachi
1
Naohiro Adachi
1
Yasunori Akagi
1
Takahito Aoto
1
Ryoma Bise
1
Daiki Chijiwa
1
Ryoji Eki
1
Yuki Endo
1
Kenji Enomoto
1
Takuma Fukuda
1
Daichi Haraguchi
1
Taku Hasegawa
1
Akio Hayakawa
1
Keigo Hirakawa
1
Yutaro Hirao
1
Tomoki Ichikawa
1
Takuya Ikeda
1
Satoshi Ikehata
1
Taichi Iki
1
Naoto Inoue
1
Masato Ishii
1
Takashi Isobe
1
Yusuke Iwasawa
1
Shun Iwase
1
Ryota Kanai
1
Sekitoshi Kanai
1
Yoshihiro Kanamori
1
Takuhiro Kaneko
1
Yusuke Kato
1
Jiro Katto
1
Kazuhiko Kawamoto
1
Yuki Kawana
1
Hiroshi Kera
1
Yuta Kikuchi
1
Kazuma Kitazawa
1
Kiyoshi Kiyokawa
1
Yuya Kobayashi
1
Kazuhito Koishida
1
Taku Komura
1
Kazuki Kozuka
1
Shuhei Kurita
1
Takahiro Maeda
1
Atsuto Maki
1
Koki Mashita
1
Yuto Matsubara
1
Yusuke Matsui
1
Hidenobu Matsuki
1
Yasuyuki Matsushita
1
Hayato Mitani
1
Shinsuke Mori
1
Ryugo Morita
1
Yusuke Moriuchi
1
Riku Murai
1
Hajime Nagahara
1
Kiyohiro Nakayama
1
Kyosuke Nishida
1
Shin’ya Nishida
1
Taichi Nishimura
1
Koichi Nishiwaki
1
Sho Nitta
1
Shohei Nobuhara
1
Takeshi Noda
1
Masahiro Nomura
1
Takeru Oba
1
Hideya Ochiai
1
Fumio Okura
1
Taishi Ono
1
Hiroki Ota
1
Junji Otsuka
1
Kuniko Saito
1
Yoichi Sato
1
Shintaro Shiba
1
Yoshihisa Shibata
1
Wataru Shimoda
1
Hirotaka Shinozaki
1
Takahiro Shirakawa
1
Jun Suzuki
1
Tsuyoshi Takatani
1
Takafumi Taketomi
1
Hiromu Taketsugu
1
Ryota Tanaka
1
Shinya Tanaka
1
Riku Togashi
1
Satoshi Tsutsui
1
Hideaki Uchiyama
1
Takeshi Uemori
1
Norimichi Ukita
1
Naoto Usuyama
1
Hiromi Wakaki
1
Ko Watanabe
1
Kota Yamaguchi
1
Shin’ya Yamaguchi
1
Kohei Yamashita
1
Naoto Yokoya
1
Tomoya Yoshida
1
Taiki Yoshino
1
Toshiya Yura
1

図2　CVPR2025　日本人研究者個人別採択件数

CVPR2018-2025 累積採択数
日本人著者ランキング
著者 採択数
Yasuyuki Matsushita
26
Tatsuya Harada
24
Yoichi Sato
22
Yasutaka Furukawa
20
Shunsuke Saito
20
Imari Sato
17
Ko Nishino
15
Kiyoharu Aizawa
11
Takuhiro Kaneko
10
Koki Nagano
10
Shohei Nobuhara
10
Toshihiko Yamasaki
10
Yasuhiro Mukaigawa
9
Yuta Nakashima
9
Masayoshi Tomizuka
9
Yoshitaka Ushiku
9
Takayuki Okatani
8
Masatoshi Okutomi
8
Mayu Otani
8
Hirokatsu Kataoka
7
Takumi Kobayashi
7
Hajime Nagahara
7
Fumio Okura
7
Norimichi Ukita
7
Yasushi Yagi
7
Kenichiro Tanaka
6
Akihiko Torii
6
Yoshimitsu Aoki
5
Hidekata Hontani
5
Daiki Ikami
5
Satoshi Ikehata
5
Naoto Inoue
5
Takashi Isobe
5
Taku Komura
5
Yusuke Sugano
5
Tsuyoshi Takatani
5
Kota Yamaguchi
5
Ryoma Bise
4
Takuya Funatomi
4
Ryo Hachiuma
4
Yusuke Hirota
4
Tomoki Ichikawa
4
Katsushi Ikeuchi
4
Takeo Kanade
4
Jiro Katto
4
Ryo Kawahara
4
Hiroyuki Kubo
4
Yuki Mitsufuji
4
Yusuke Moriuchi
4
Hideki Nakayama
4
Takahiro Okabe
4
Kuniaki Saito
4
Shin’ichi Satoh
4
Akihiro Sugimoto
4
Tatsunori Taniai
4
Riku Togashi
4
Takahito Aoto
3
Yuta Asano
3
Kenji Enomoto
3
Ryuhei Hamaguchi
3
Go Irie
3
Hiroshi Ishikawa
3
Mariko Isogawa
3
Hiroharu Kato
3
Kotaro Kikuchi
3
Norimasa Kobori
3
Kazuki Kozuka
3
Yasushi Makihara
3
Takeshi Naemura
3
Kazuki Okami
3
Taishi Ono
3
Fumihiko Sakaue
3
Ken Sakurada
3
Hiroaki Santo
3
Jun Sato
3
Yutaka Satoh
3
Taiki Sekii
3
Takaaki Shiratori
3
Shoichiro Takeda
3
Takafumi Taketomi
3
Towaki Takikawa
3
Rin-ichiro Taniguchi
3
Keisuke Tateno
3
Seiichi Uchida
3
Takeshi Uemori
3
Takuma Yagi
3
Takayoshi Yamashita
3
Rio Yokota
3
Tatsuya Yokota
3
Naoto Yokoya
3
Ryo Yonetani
3
Osamu Yoshie
3

図3　CVPR2018-2025 累積採択数 日本人著者ランキング

■論文出現キーワードの推移

最後に、論文タイトルに含まれるキーワードの比率も、調査いたしました（表3）。

今年も昨年同様に、“diffusion（拡散モデル）”および“language model（言語モデル）”がトレンドキーワードとして挙がっていることが分かります。なお、今回利用したキーワードリストにはdiffusionと組み合わせたものがなかったため抽出はできていませんが、“diffusion transformer”などの出現頻度も高く、拡散モデルにおけるtransformerの利用も一般的になりつつあります。vision-language modelなど、言語モデルとの接続や活用も非常に重要になってきていると言えます。multi modalなど複数のモダリティを統合的に扱う研究の重要性も増大しており、近年のAIトレンドの一端が見て取れます。

また、今年の顕著な特徴としてGaussian splattingが躍進していることが挙げられます。NeRFに変わる軽量な三次元レンダリング方法として脚光を浴びて以来、研究として深化しながら標準的なツールとして根付きつつあるのが見て取れます。学習や推論の効率性に関わるキーワード（efficientやdistillation）も増えてきており、今後AIの普及や実用化につれて重要になってくると思われます。

＊ 今回より、出現キーワードの抽出方法を大きく変更いたしました。
　 過去に公開したコラム内の出現キーワードとの順位相違や、新たなキーワードが追加されております。

＊「***」がついている単語は、品詞の変化などを含めて解析しております。
　例）generat*** ⇒ generative, generation, etc…

Year 2025 2024 2023 2022 2021 2020 2019 2018 2017 2016 2015 2014
#Papers 2,872 2,716 2,357 2,063 1,660 1,466 1,294 979 783 643 602 540
Ranking Term Frequency Term Frequency Term Frequency Term Frequency Term Frequency Term Frequency Term Frequency Term Frequency Term Frequency Term Frequency Term Frequency Term Frequency
1 image 13.34% diffusion 12.04% image 13.30% image 13.11% image 12.05% image 13.64% image 14.76% image 14.40% image 15.07% image 15.09% image 17.77% image 15.56%
2 diffusion 11.98% image 11.86% video 8.80% transformer 9.50% video 8.31% video 7.37% video 7.42% video 9.30% video 8.69% video 8.55% video 7.81% segment*** 7.04%
3 video 11.73% generat*** 9.02% transformer 6.38% video 8.97% generat*** 5.30% graph 5.25% adversarial 5.80% adversarial 6.44% detect*** 5.88% detect*** 6.07% estimat*** 6.15% track 5.93%
4 generat*** 11.04% video 8.73% generat*** 5.87% detect*** 5.11% represent 5.30% object detec*** 4.84% generat*** 5.26% generat*** 5.72% segment*** 4.73% recogni*** 5.13% detect*** 6.15% video 5.37%
5 language model 5.64% efficien*** 4.82% represent 5.53% object detec*** 5.06% object detec*** 5.24% estimat*** 4.57% detect*** 5.02% feature 4.80% feature 3.96% feature 4.98% feature 5.32% recogni*** 5.37%
6 efficien*** 5.12% segment*** 4.82% object detec*** 4.72% represent 4.97% unsupervised 4.64% represent 4.57% feature 4.25% detect*** 4.29% generat*** 3.96% segment*** 4.82% segment*** 5.15% feature 5.19%
7 adapt 4.77% adapt 4.82% efficien*** 4.55% generat*** 4.77% adapt 4.58% segment*** 4.37% graph 4.25% segment*** 4.09% recogni*** 3.83% structur*** 4.82% recogni*** 4.65% detect*** 5.00%
8 multimodal 4.32% represent 4.09% adapt 4.29% graph 3.76% detect*** 4.22% adversarial 4.23% represent 4.10% object detec*** 3.78% represent 3.45% estimat*** 4.67% track 3.82% match 4.81%
9 segment*** 3.69% transformer 4.09% diffusion 4.21% self supervised 3.71% segment*** 4.04% attention 4.09% unsupervised 4.10% estimat*** 3.57% classifi*** 2.94% classifi*** 4.35% represent 3.65% estimat*** 4.44%
10 represent 3.59% synthe 3.68% detect*** 3.70% segment*** 3.62% graph 3.92% detect*** 3.96% segment*** 4.02% attention 3.47% camera 2.94% predict 3.89% match 3.49% classifi*** 4.07%
11 detect*** 3.41% detect*** 3.54% point cloud 3.49% efficien*** 3.52% feature 3.74% adapt 3.68% object detec*** 3.86% re identification 3.06% structur*** 2.81% object detec*** 3.58% efficien*** 3.49% structur*** 3.70%
12 gaussian splat 3.31% language model 3.39% reconstruct*** 3.36% adapt 3.18% efficien*** 3.43% generat*** 3.62% adapt 3.55% match 3.06% efficien*** 2.81% label 3.42% spars 3.49% reconstruct*** 3.52%
13 reconstruct*** 3.27% prompt 3.28% synthe 3.32% semantic segmentation 3.04% self supervised 3.19% recogni*** 3.62% attention 3.48% depth 3.06% depth 2.68% track 3.42% graph 3.49% label 3.52%
14 transformer 3.20% reconstruct*** 2.98% semantic segmentation 3.23% point cloud 3.04% point cloud 2.89% unsupervised 3.55% efficien*** 3.32% structur*** 3.06% match 2.68% match 3.27% structur*** 3.32% graph 3.33%
15 synthe 3.06% generaliz*** 2.80% segment*** 3.15% estimat*** 2.99% adversarial 2.83% feature 3.34% recogni*** 3.09% unsupervised 3.06% estimat*** 2.55% spars 3.27% depth 3.32% represent 3.33%
16 align 2.93% object detec*** 2.76% graph 3.06% synthe 2.89% reconstruct*** 2.83% predict 3.27% estimat*** 3.01% predict 2.96% caption 2.55% represent 2.96% camera 3.32% depth 3.15%
17 dataset 2.93% interact 2.69% self supervised 3.02% attention 2.84% label 2.83% search 3.00% predict 2.94% recogni*** 2.96% predict 2.43% efficien*** 2.96% classifi*** 3.16% spars 2.96%
18 graph 2.75% multimodal 2.65% feature 2.97% adversarial 2.75% recogni*** 2.71% efficien*** 2.86% structur*** 2.78% track 2.86% localiz*** 2.43% salien 2.64% salien 2.99% camera 2.96%
19 interact 2.72% feature 2.58% estimat*** 2.64% predict 2.70% depth 2.71% reconstruct*** 2.86% point cloud 2.63% localiz*** 2.86% weakly supervised 2.43% reconstruct*** 2.49% descript 2.99% efficien*** 2.78%
20 feature 2.68% dataset 2.58% dataset 2.55% feature 2.70% estimat*** 2.71% synthe 2.73% depth 2.40% represent 2.76% context 2.30% graph 2.49% label 2.99% manifold 2.59%
21 benchmark 2.65% estimat*** 2.43% generaliz*** 2.51% unsupervised 2.60% synthe 2.59% point cloud 2.59% transfer 2.32% synthe 2.76% attention 2.30% camera 2.49% retriev*** 2.66% hierarch 2.59%
22 prompt 2.61% align 2.36% unsupervised 2.47% label 2.56% predict 2.41% depth 2.52% match 2.24% weakly supervised 2.76% label 2.30% localiz*** 2.33% cluster 2.49% localiz*** 2.41%
23 consisten*** 2.61% consisten*** 2.25% predict 2.42% align 2.46% search 2.41% self supervised 2.39% pose estimat*** 2.24% graph 2.35% graph 2.17% descript 2.18% object detec*** 2.49% kernel 2.41%
24 estimat*** 2.61% edit 2.25% adversarial 2.42% recogni*** 2.46% semantic segmentation 2.35% structur*** 2.39% reconstruct*** 2.16% pose estimat*** 2.25% adversarial 2.17% dataset 2.18% search 2.33% constrain 2.22%
25 generaliz*** 2.47% text to image 2.21% recogni*** 2.34% few shot 2.36% attention 2.29% context 2.18% synthe 2.16% efficien*** 2.25% regress 2.17% search 2.18% predict 2.33% predict 2.04%
26 edit 2.40% predict 2.17% label 2.25% dataset 2.36% transformer 2.23% label 2.11% context 2.16% context 2.15% unsupervised 2.17% hierarch 2.02% reconstruct*** 2.33% search 2.04%
27 predict 2.33% enhance 2.14% localiz*** 2.08% reconstruct*** 2.27% domain adaptation 2.23% transfer 2.05% semantic segmentation 2.09% reconstruct*** 2.04% object detec*** 2.17% linear 1.87% hierarch 2.33% pose estimat*** 2.04%
28 prior 2.23% fusion 2.14% consisten*** 2.04% match 2.22% few shot 2.23% domain adaptation 2.05% domain adaptation 2.01% semantic segmentation 2.04% action recogni***tion 2.17% constrain 1.87% constrain 2.33% parsing 2.04%
29 enhance 2.12% distill 2.10% semi supervised 1.96% depth 2.22% camera 2.17% classifi*** 1.98% re identification 1.93% camera 1.94% track 2.04% unsupervised 1.87% align 2.16% 3d reconstruction 1.85%
30 zero shot 2.09% unsupervised 2.10% prompt 1.96% semi supervised 2.03% localiz*** 2.17% match 1.98% embedding 1.93% transfer 1.84% adapt 2.04% depth 1.87% manifold 2.16% cluster 1.85%
31 attention 2.05% prior 2.06% distill 1.91% consisten*** 2.03% semi supervised 2.11% pose estimat*** 1.98% hierarch 1.86% classifi*** 1.84% pose estimat*** 1.92% pose estimat*** 1.87% adapt 2.16% context 1.85%
32 point cloud 1.98% graph 2.03% attention 1.91% weakly supervised 1.98% super resolution 1.99% semantic segmentation 1.91% dataset 1.78% adapt 1.84% semantic segmentation 1.92% align 1.71% localiz*** 1.99% action recogni***tion 1.85%
33 spars 1.98% point cloud 1.99% pre train 1.91% interact 1.98% structur*** 1.93% track 1.91% track 1.78% question answer 1.74% embedding 1.79% re identification 1.71% regulariz 1.99% align 1.67%
34 depth 1.98% semantic segmentation 1.92% few shot 1.87% contrastive learning 1.98% track 1.87% interact 1.91% search 1.70% end to end 1.74% in the wild 1.66% action recogni***tion 1.71% kernel 1.83% unsupervised 1.67%
35 object detec*** 1.92% adversarial 1.88% align 1.83% structur*** 1.93% align 1.81% dataset 1.84% weakly supervised 1.70% point cloud 1.63% synthe 1.66% embedding 1.56% prior 1.83% scalab 1.67%
36 multi modal 1.85% zero shot 1.84% prior 1.79% localiz*** 1.93% transfer 1.81% attack 1.84% caption 1.70% geometry 1.63% retriev*** 1.66% prior 1.56% discover 1.83% salien 1.48%
37 video generation 1.85% attention 1.80% match 1.79% spars 1.93% consisten*** 1.81% hierarch 1.77% label 1.62% domain adaptation 1.63% descript 1.66% cluster 1.56% action recogni***tion 1.83% bayes 1.48%
38 compress 1.74% camera 1.77% hierarch 1.70% classifi*** 1.83% match 1.75% few shot 1.77% end to end 1.55% embedding 1.53% re identification 1.66% regress 1.56% subspace 1.66% linear 1.48%
39 distill 1.67% multi modal 1.77% spars 1.70% transfer 1.78% pose estimat*** 1.75% semi supervised 1.70% camera 1.55% constrain 1.53% residual 1.66% semantic segmentation 1.56% regress 1.66% transfer 1.48%
40 text to image 1.67% match 1.73% structur*** 1.66% pose estimat*** 1.74% re identification 1.69% super resolution 1.57% classifi*** 1.47% salien 1.53% composition 1.66% transfer 1.40% dataset 1.66% online 1.48%
41 composition 1.67% spars 1.69% weakly supervised 1.66% generaliz*** 1.64% weakly supervised 1.69% re identification 1.57% localiz*** 1.47% spars 1.53% zero shot 1.66% weakly supervised 1.40% context 1.66% dictionary 1.48%
42 camera 1.64% benchmark 1.66% pose estimat*** 1.62% end to end 1.64% dataset 1.63% camera 1.57% retriev*** 1.47% caption 1.53% salien 1.41% online 1.40% low rank 1.49% adapt 1.30%
43 adversarial 1.57% pre train 1.62% super resolution 1.57% context 1.64% classifi*** 1.57% compress 1.57% zero shot 1.47% super resolution 1.43% dataset 1.41% composition 1.24% image classifi*** 1.33% group 1.30%
44 fusion 1.57% label 1.62% render 1.57% compress 1.59% compress 1.45% end to end 1.50% compress 1.47% compress 1.43% spatio temporal 1.41% kernel 1.24% fusion 1.33% probabilistic 1.30%
45 structur*** 1.53% structur*** 1.55% zero shot 1.53% attack 1.59% end to end 1.45% weakly supervised 1.43% salien 1.47% interact 1.43% deep neural network 1.41% adapt 1.24% super resolution 1.33% object detec*** 1.30%
46 render 1.50% recogni*** 1.55% transfer 1.49% pre train 1.54% embedding 1.39% reasoning 1.43% few shot 1.39% composition 1.43% question answer 1.28% group 1.24% generat*** 1.33% regress 1.30%
47 self supervised 1.50% context 1.55% interact 1.49% hierarch 1.49% instance segmentation 1.39% generaliz*** 1.30% fusion 1.39% zero shot 1.43% subspace 1.28% image classifi*** 1.24% unsupervised 1.16% subspace 1.30%
48 mamba 1.50% self supervised 1.51% retriev*** 1.44% prior 1.49% interact 1.39% embedding 1.30% self supervised 1.39% align 1.33% align 1.28% recover 1.24% metric learning 1.16% synthe 1.30%
49 foundation model 1.46% localiz*** 1.51% edit 1.44% camera 1.45% distill 1.33% salien 1.30% regulariz 1.39% dataset 1.33% reconstruct*** 1.28% question answer 1.09% linear 1.16% evaluat*** 1.11%
50 reasoning 1.46% render 1.47% depth 1.44% search 1.40% prior 1.33% in the wild 1.30% interact 1.31% fusion 1.23% encod*** 1.28% attention 1.09% transfer 1.16% interact 1.11%
51 latent 1.43% few shot 1.44% fusion 1.40% cross modal 1.25% generaliz*** 1.27% cluster 1.30% align 1.31% action recogni***tion 1.23% regulariz 1.28% zero shot 1.09% semantic segmentation 1.16% non rigid 1.11%
52 label 1.43% tuning 1.40% track 1.36% super resolution 1.25% contrastive 1.27% prior 1.23% consisten*** 1.31% regress 1.23% search 1.28% consisten*** 1.09% re identification 1.16% descript 1.11%
53 tuning 1.39% depth 1.40% pretrain 1.36% track 1.25% retriev*** 1.27% instance segmentation 1.23% action recogni***tion 1.24% memor 1.23% online 1.28% interact 1.09% face recogni*** 1.00% composition 1.11%
54 retriev*** 1.36% super resolution 1.36% contrastive 1.32% multimodal 1.21% context 1.27% regulariz 1.23% reasoning 1.24% disentangl 1.12% multimodal 1.15% context 1.09% material 1.00% recover 1.11%
55 restor 1.32% latent 1.36% camera 1.32% render 1.21% hierarch 1.21% group 1.23% regress 1.24% conditional 1.12% hierarch 1.15% parsing 1.09% pose estimat*** 1.00% color 1.11%
56 token 1.29% hierarch 1.33% contrastive learning 1.28% contrastive 1.16% attack 1.21% spars 1.09% instance segmentation 1.24% cluster 1.12% linear 1.15% manifold 1.09% rgbd 1.00% in the wild 1.11%
57 semantic segmentation 1.29% composition 1.33% enhance 1.28% domain adaptation 1.16% in the wild 1.21% localiz*** 1.09% super resolution 1.24% in the wild 1.02% image classifi*** 1.15% mining 1.09% parsing 1.00% enhance 1.11%
58 few shot 1.29% retriev*** 1.33% benchmark 1.23% retriev*** 1.16% cross modal 1.21% fusion 1.09% question answer 1.24% manifold 1.02% color 1.15% deep neural network 1.09% consisten*** 1.00% weakly supervised 0.93%
59 geometry 1.25% track 1.29% attack 1.23% in the wild 1.11% disentangl 1.21% composition 1.02% composition 1.24% consisten*** 1.02% constrain 1.15% light field 1.09% embedding 1.00% semi supervised 0.93%
60 match 1.25% pose estimat*** 1.25% regulariz 1.15% re identification 1.06% spars 1.21% translat 1.02% generaliz*** 1.16% kernel 1.02% cross modal 1.15% synthe 1.09% mining 1.00% annotat 0.93%
61 attack 1.25% transfer 1.25% compress 1.15% cluster 1.06% render 1.21% memor 1.02% cluster 1.16% label 1.02% spars 1.15% latent 0.93% convex 1.00% image classifi*** 0.93%
62 unsupervised 1.22% foundation model 1.14% composition 1.15% embedding 1.06% composition 1.21% disentangl 1.02% metric learning 1.08% reinforcement learning 1.02% low rank 1.02% annotat 0.93% generaliz*** 1.00% prior 0.93%
63 manipulat 1.22% gaussian splat 1.14% cross modal 1.15% online 1.06% latent 1.21% consisten*** 1.02% attack 1.08% translat 1.02% parsing 1.02% retriev*** 0.93% composition 1.00% outlier 0.93%
64 super resolution 1.22% attack 1.11% multimodal 1.15% distill 1.06% fusion 1.15% cross modal 1.02% spars 1.01% hierarch 0.92% benchmark 1.02% 3d reconstruction 0.93% correlation 1.00% generaliz*** 0.93%
65 medic 1.18% medic 1.07% latent 1.11% regulariz 1.06% reasoning 1.08% online 0.96% memor 1.01% correlation 0.92% autoencod***er 1.02% caption 0.78% bayes 0.83% covarian 0.93%
66 image generation 1.18% lidar*** 1.07% geometry 1.11% instance segmentation 1.01% salien 1.02% zero shot 0.96% translat 1.01% latent 0.92% bayes 1.02% generat*** 0.78% coarse to fine 0.83% object tracking 0.93%
67 track 1.15% autonomous 1.03% lidar*** 1.11% zero shot 1.01% translat 1.02% action recogni***tion 0.96% in the wild 1.01% self supervised 0.92% domain adaptation 1.02% encod*** 0.78% deep neural network 0.83% retriev*** 0.93%
68 localiz*** 1.15% instruction 1.03% bias 1.11% composition 1.01% benchmark 1.02% align 0.96% disentangl 1.01% deep neural network 0.92% manifold 1.02% ranking 0.78% benchmark 0.83% geometry 0.74%
69 context 1.11% bias 0.99% context 1.06% domain generaliz*** 1.01% cluster 1.02% deep neural network 0.89% benchmark 0.93% retriev*** 0.92% end to end 1.02% convex 0.78% light field 0.83% calibration 0.74%
70 pre train 1.11% test time 0.96% embedding 1.06% multi modal 1.01% object tracking 1.02% constrain 0.89% prior 0.93% residual 0.92% transfer 0.89% color 0.78% weakly supervised 0.83% object recogni*** 0.74%
71 robot 1.11% weakly supervised 0.92% multi modal 1.06% fusion 1.01% probabilistic 0.96% parsing 0.89% cross modal 0.93% style transfer 0.82% light field 0.89% benchmark 0.78% in the wild 0.83% convex 0.74%
72 semi supervised 1.08% semi supervised 0.92% classifi*** 1.02% memor 1.01% contrastive learning 0.90% render 0.89% semi supervised 0.93% group 0.82% correlation 0.89% super resolution 0.78% online 0.83% ranking 0.74%
73 collaborat 1.08% encod*** 0.88% domain generaliz*** 0.98% lidar*** 1.01% caption 0.84% retriev*** 0.89% annotat 0.93% regulariz 0.82% reinforcement learning 0.89% fusion 0.78% 3d reconstruction 0.83% pedestrian detection 0.74%
74 recogni*** 1.08% memor 0.88% domain adaptation 0.98% benchmark 0.96% face recogni*** 0.84% conditional 0.89% geometry 0.85% prior 0.82% kernel 0.89% in the wild 0.78% annotat 0.83% extract 0.74%
75 regress 1.08% geometry 0.88% cluster 0.98% enhance 0.96% panoptic segmentation 0.84% edit 0.89% image classifi*** 0.85% quantiz 0.82% calibration 0.89% end to end 0.78% compress 0.83% regulariz 0.74%
76 grounding 1.05% regress 0.88% quantiz 0.98% disentangl 0.96% memor 0.84% latent 0.82% latent 0.85% descript 0.72% metric learning 0.89% probabilistic 0.62% domain adaptation 0.66% minimiz 0.74%
77 test time 1.05% image generation 0.85% language model 0.98% geometry 0.96% bias 0.84% color 0.82% face recogni*** 0.85% intrinsic 0.72% cluster 0.89% dictionary 0.62% sensor 0.66% fusion 0.74%
78 hierarch 1.05% grounding 0.85% grounding 0.94% latent 0.96% interpret 0.84% enhance 0.82% constrain 0.85% reasoning 0.72% fusion 0.89% compress 0.62% recover 0.66% multimodal 0.74%
79 anomaly detection 1.01% compress 0.85% instance segmentation 0.89% restor 0.96% restor 0.84% caption 0.82% discover 0.85% face recogni*** 0.72% 3d reconstruction 0.89% discover 0.62% non rigid 0.66% encod*** 0.74%
80 encod*** 1.01% token 0.81% action recogni***tion 0.85% knowledge distillat 0.96% pre train 0.78% multi modal 0.82% encod*** 0.85% 3d reconstruction 0.72% recover 0.89% evaluat*** 0.62% multiple instance learning 0.66% low rank 0.74%
81 cross modal 0.98% boosting 0.81% encod*** 0.85% caption 0.87% collaborat 0.78% reinforcement learning 0.82% descript 0.85% light field 0.72% annotat 0.89% metric learning 0.62% defocus 0.66% re identification 0.56%
82 lidar*** 0.98% reasoning 0.81% frequency 0.81% translat 0.87% regulariz 0.78% face recogni*** 0.82% residual 0.85% parsing 0.61% face recogni*** 0.77% object tracking 0.62% encod*** 0.66% boosting 0.56%
83 evaluat*** 0.98% embedding 0.81% disentangl 0.81% grounding 0.87% grounding 0.72% regress 0.82% spatio temporal 0.77% linear 0.61% consisten*** 0.77% regulariz 0.62% parametric 0.66% generat*** 0.56%
84 autonomous 0.87% collaborat 0.81% manipulat 0.81% medic 0.87% spatio temporal 0.72% distill 0.75% parsing 0.77% mining 0.61% interact 0.77% non rigid 0.62% semi supervised 0.66% semantic segmentation 0.56%
85 fine tun*** 0.87% federated learning 0.77% restor 0.81% navigat 0.82% discover 0.66% encod*** 0.75% deep neural network 0.77% multimodal 0.61% dynamics 0.77% quantiz 0.62% correction 0.66% statistical 0.56%
86 animation 0.84% disentangl 0.77% knowledge distillat 0.81% group 0.82% conditional 0.66% transformer 0.75% quantiz 0.77% interpret 0.61% super resolution 0.77% multimodal 0.62% restor 0.66% frequency 0.56%
87 memor 0.80% restor 0.77% text to image 0.77% bias 0.82% lidar*** 0.66% spatio temporal 0.75% conditional 0.77% online 0.61% intrinsic 0.77% conditional 0.62% calibration 0.66% dataset 0.56%
88 scalab 0.80% navigat 0.77% autoencod***er 0.77% regress 0.82% regress 0.66% geometry 0.68% linear 0.77% probabilistic 0.61% prior 0.77% calibration 0.62% spatio temporal 0.56%
89 pose estimat*** 0.80% cluster 0.74% discover 0.77% incremental learning 0.82% multimodal 0.66% image classifi*** 0.68% kernel 0.70% encod*** 0.61% one shot 0.64% memor 0.62% compress 0.56%
90 classifi*** 0.77% domain generaliz*** 0.74% annotat 0.77% quantiz 0.77% action recogni***tion 0.66% knowledge distillat 0.68% group 0.70% answer 0.61% latent 0.64% subspace 0.62% random forest 0.56%
91 state space model 0.77% frequency 0.74% recover 0.77% annotat 0.77% edit 0.66% autonomous 0.68% distill 0.70% scalab 0.51% instance segmentation 0.64% face recogni*** 0.62% quantiz 0.56%
92 online 0.77% anomaly detection 0.74% in the wild 0.77% action recogni***tion 0.77% online 0.66% medic 0.61% multimodal 0.70% non rigid 0.51% discover 0.64% metric learning 0.56%
93 frequency 0.77% domain adaptation 0.74% token 0.77% federated learning 0.72% encod*** 0.60% quantiz 0.61% color 0.70% benchmark 0.51% style transfer 0.64% domain adaptation 0.56%
94 agent 0.77% online 0.74% test time 0.72% probabilistic 0.72% zero shot 0.60% residual 0.61% collaborat 0.70% ranking 0.51% conditional 0.64% non linear 0.56%
95 disentangl 0.73% constrain 0.74% continual learning 0.72% edit 0.72% mining 0.60% bias 0.61% navigat 0.62% perceptual 0.51% memor 0.64% articulated 0.56%
96 3d reconstruction 0.73% contrastive learning 0.74% memor 0.72% object tracking 0.72% style transfer 0.60% multimodal 0.61% autoencod***er 0.54% recover 0.51% quantiz 0.64% diffusion 0.56%
97 interpret 0.73% end to end 0.74% anomaly detection 0.68% image generation 0.68% black box 0.60% object tracking 0.61% inpaint 0.54% attack 0.51% group 0.64% relaxation 0.56%
98 in the wild 0.73% 3d reconstruction 0.70% tuning 0.68% correlation 0.63% one shot 0.60% panoptic segmentation 0.55% decod 0.54% manipulat 0.51% minimiz 0.64%
99 transfer 0.70% regulariz 0.70% constrain 0.68% reasoning 0.63% recover 0.60% meta learning 0.55% autonomous 0.54% one shot 0.51% geometry 0.64%
100 bias 0.70% scalab 0.70% conditional 0.68% diffusion 0.63% correction 0.60% benchmark 0.55% enhance 0.54% subspace 0.51% reasoning 0.51%
101 cluster 0.66% discover 0.70% online 0.68% discover 0.63% quantiz 0.60% prun 0.55% object tracking 0.54% enhance 0.51% agent 0.51%
102 physics 0.66% prun 0.66% federated learning 0.64% boosting 0.63% navigat 0.60% lidar*** 0.55% correlation 0.54% restor 0.51% point cloud 0.51%
103 instruction 0.66% calibration 0.66% image classifi*** 0.64% manipulat 0.63% geometry 0.60% descript 0.55% style transfer 0.54% discover 0.51% self supervised 0.51%
104 weakly supervised 0.63% group 0.66% scalab 0.64% encod*** 0.58% annotat 0.60% image generation 0.55% correction 0.54% prun 0.51% interpret 0.51%
105 constrain 0.63% video generation 0.63% salien 0.64% 3d reconstruction 0.58% inpaint 0.54% grounding 0.54% grounding 0.51% fine tun*** 0.51%
106 incremental learning 0.63% caption 0.63% caption 0.64% evaluat*** 0.58% topolog 0.54% prun 0.54% object recogni*** 0.51%
107 preference 0.63% cross modal 0.63% reasoning 0.64% correction 0.53% evaluat*** 0.54% manifold 0.54% ranking 0.51%
108 quantiz 0.63% evaluat*** 0.63% backdoor 0.64% token 0.53% perceptual 0.54% evaluat*** 0.51%
109 search 0.63% classifi*** 0.63% vision and language 0.64% vision and language 0.53% incremental learning 0.54% compress 0.51%
110 discover 0.59% manipulat 0.63% medic 0.64% equivarian 0.53% dynamics 0.54% generaliz*** 0.51%
111 embod 0.59% decod 0.59% autonomous 0.60% question answer 0.53% enhance 0.54% object tracking 0.51%
112 simulation 0.59% quantiz 0.59% navigat 0.60% image classifi*** 0.53% group 0.54% medic 0.51%
113 heterogeneous 0.59% continual learning 0.59% end to end 0.60% continual learning 0.53% domain generaliz*** 0.54%
114 boosting 0.59% image classifi*** 0.59% regress 0.60% panoptic segmentation 0.53% constrain 0.54%
115 out of distribution 0.59% interpret 0.55% translat 0.55% noisy label 0.53%
116 federated learning 0.59% segment*** anything 0.55% 3d reconstruction 0.55%
117 embedding 0.56% pretrain 0.55% boosting 0.55%
118 color 0.56% knowledge distillat 0.55% descript 0.55%
119 domain generaliz*** 0.56% speech 0.52% re identification 0.55%
120 prun 0.52% instance segmentation 0.52% image generation 0.51%
121 recover 0.52% descript 0.52% out of distribution 0.51%
122 re identification 0.52% search 0.51%
123 spatio temporal 0.52% kernel 0.51%
124 embod 0.52% open set 0.51%
125 explanation 0.51%

表3　論文出現キーワード推移（2014-2025年）

■まとめ

今年も、CVPR2025開始前の予習的な意味合いで統計情報を先行してまとめ公開いたしました（2025年6月4日情報解析）。今週より始まるCVPR2025に現地・オンラインで参加される方の参考になれば嬉しい限りです。
　

● ResearchPortメンバーシップ募集：
https://research-p.com/?memberpage=registration
ResearchPortでは、研究者・技術者の方の研究事業開発サポートやキャリアサポートサービスを提供しております。ご興味がある方はResearchPortメンバーシップへご登録下さい。

　
編集：ResearchPort事業部

■Contact

本記事に関する質問や、ご意見・ご要望などがございましたらResearchPortまでお問い合わせください。
https://research-p.com/contactform/

computer vision CVPR トップカンファレンス定点観測
関連記事
メンバーシップ登録

Membership Registration

博士課程在籍中の方 ポスドクの方 大学教員の方 企業研究者/技術者

ResearchPortでは、研究者・技術者の皆さま一人ひとりのご希望に寄り添い、キャリア相談や副業案件、技術顧問依頼、研究プロジェクトのご紹介を個別に行っています。また、研究成果の技術広報支援や研究パートナー探しなど、キャリアと研究活動の両面をサポートいたします。
ご興味のある方は、ぜひResearchPortメンバーシップにご登録ください。