Imanol Miranda
@I_MirandaM
Followers
40
Following
89
Media
9
Statuses
62
PhD student at HiTZ Zentroa (@Hitz_zentroa) / IXA Group (@IxaTaldea) and the University of Basque Country (@txiomatika).
Joined November 2011
RT @Hitz_zentroa: Ayer uno de nuestros investigadores, Oscar Sainz (@osainz59), fue galardonado con el premio a la mejor tesis doctoral en….
0
9
0
RT @HABE_erakundea: “AA eta euskara: tresnak eta garapenak (II)“ Oier Lopez de Lacalle Lekuona @Hitz_zentroa @Irakasbil @HabeLiburutegia @u….
0
10
0
RT @Hitz_zentroa: HiTZ zentroko bilera orokorra egin dugu aste honetan Iruñean . We held the general meeting of the HiTZ center this week i….
0
7
0
RT @Hitz_zentroa: In last week’s seminar session @anaryegen talked about Mining Argument Structures in Medical Texts (.
0
3
0
RT @Hitz_zentroa: Ekaia EHUko Zientzia eta Teknologia aldizkariak Adimen Artifizialeko ale berezia argitaratu berri du. Bertan, @Hitz_zentr….
0
4
0
RT @berria: «Kaixo, Latxa naiz. Zer jakin nahi duzu gaur?». Euskarazko txatbota sortu du EHUko HiTZ ikerketa zentroak. Garatzaileek eta enp….
0
11
0
Key takeaway: Adding simple structure at inference-time, through image crops and text segments, is a powerful, training-free way to improve Vision-Language Compositionality performance. Joint work with @AnderSala @eagirre @gazkune @IxaGroup @Hitz_zentroa.
0
1
3
Our analysis shows that:.1. There is room to improve the quality of extracted text segments. 2. Our method achieves significant performance gains in Winoground's non-trivial instances. 3. Isolated image crops can lose size and quantity information, leaving room for improvement.
1
1
4
Why are image crops crucial? 🤔 We found that simply adding text segments isn't enough. The biggest performance gains come when text segments are paired with image crops, proving the power of serial image computing.
1
1
3
We've evaluated it across three diverse datasets: BiVLC, Winoground (171 instances), and BiSCoR-Ctrl. See the significant improvements by inference-time approach (ITA) on three existing models:
1
1
3
Our approach is straightforward yet effective:.1. Divide the image into smaller crops. 2. Extract text segments capturing objects, attributes and relations. 3. Use the VLM to find image crops that best fit the text segments. 4. Aggregate matching similarities for the final score.
1
1
4
#newHitzPaper.Can a simple inference-time approach unlock better Vision-Language Compositionality?🤯.Our latest paper shows how adding structure at inference significantly boosts performance in popular dual-encoder VLMs on different datasets. Read more:
1
6
10
RT @Hitz_zentroa: [1/7].#newHitzPaper. Many languages are underserved by open LLMs, and face the following question: Which is the best way….
0
12
0
RT @osainz59: Do you know that you can continue pretraining Instructed LLMs without losing their instruction following capabilities?. We di….
0
6
0
RT @Hitz_zentroa: We are delighted to announce the 9 papers (4 main and 5 findings) were accepted at #ACL2025NLP #ACL2025!🎉🎉🎉. Stay tuned f….
0
4
0
RT @Blanca_C_Fi: 🎉 The results of the shared task are out!.We’re thrilled with the turnout — 13 amazing teams submitted their systems to th….
0
7
0
RT @Hitz_zentroa: Aste honetan, @gazkune @EgunOnETB saioan ChatGPTek eta hizkuntza ereduek dituzten ur eta energia beharrei buruz aritu da….
0
4
0
RT @eagirre: Probably the best master on language technologies, computational linguistics and generative AI principles. Most of our PhDs co….
0
3
0