Happy to be listed as an Outstanding Reviewer again #ICCV2025 🌟

[3] Cinematic Audio Source Separation Using Visual Cues, Kang Zhang*, Suyeon Lee*, Arda Senocak+, Joon Son Chung+, #CVPR2026

[2] How Far Can We Go With Synthetic Data for Audio-Visual Sound Source Localization?, Arda Senocak*, Sooyoung Park*, Tae-Hyun Oh, Joon Son Chung

[1] Seeing Through Touch: Tactile-Driven Visual Localization of Material Regions, Seongyu Kim, Seungwoo Lee, Hyeonggon Ryu, Joon Son Chung, Arda Senocak

I have three papers (one first-author and two corresponding-author papers) accepted to #CVPR2026. These works reflect our continued efforts to push multimodal learning forward through self-supervised computer vision methods that learn from sound and touch: ⬇️

Thanks for sharing our work @_akhaliq 🤩 Code is coming very soon 🐍🎙️

``Audio Mamba: Bidirectional State Space Model for Audio Representation Learning,'' Mehmet Hamza Erol, Arda Senocak, Jiu Feng, Joon Son Chung,

PS: 👏 A big shoutout to @guy_yariv for his great work AudioToken (Interspeech 2023).

You can try our 🤗 HF demo here: https://t.co/HFPxkCNcUy @_akhaliq

Finally, We pair a single image with different object sounds, highlighting our method's interactive sound localization power 🎶

We qualitatively compared our method to a text-conditioned open-world segmentation model🧐 Results suggest that sound sources aren't always well localized with text info🤔 But our audio-visual correspondence-based model excels in pinpointing sounding objects! 🎯

Our method shines in extensive experiments, surpassing state-of-the-art approaches by a wide margin! 🚀 🏆 Check out these qualitative results too! Our model produces precise, compact localization maps for sounding objects 🎯

Our method: 1- Converts audio to CLIP-compatible tokens for audio-driven embeddings. 2- Creates audio-grounded masks for the audio embeddings. 3- Extracts image features from highlighted regions, aligning them with audio embeddings using audio-visual correspondence.

Can Foundational Model Help Alignment? 🤔 We aimed to use CLIP's robust multi-modal alignment into audio-visual correspondence🌟But without using any explicit text input, just pure audio-visual correspondence!

Introducing our new #WACV2024 paper!🎉 📝 : https://t.co/JbFyRbRidN 🤗@huggingface Demo: https://t.co/HFPxkCNKK6 Wed. 5th (Today) 8:00PM-10:00PM "Can CLIP Help Sound Source Localization?"

To conclude: “Sound Source Localization is All About Cross-Modal Alignment!” If you're at #ICCV2023, don't miss our poster presentation! We're eager to discuss and answer your questions :)

There's one more goodie in the paper. By synthetically pairing a single image with various sounds from objects present in a scene, we showcase our method's strength in interactive sound localization. We observe a clear edge over competing methods!

Cross-modal semantic alignment is important in understanding semantically mismatched audio-visual events, e.g., silent objects, offscreen sounds. Our method performs better than the competing methods in false positive detection as this task also require cross-modal interaction.

We put all existing sound localization methods to the test in the cross-modal retrieval task. Thanks to our robust cross-modal alignment, we outshine other state-of-the-art methods 🌟 High sound localization performance doesn't always translate to superior cross-modal retrieval!

The current evaluation settings do not capture the true sound source localization ability. We propose two auxiliary evaluation tasks stemming from the cross-modal alignment task : 1️⃣Interactive sound localization 2️⃣Cross-modal retrieval