Presenting my VLM research at the Aberdeen Uni Interdisciplinary Symposium was a success!
However, the empirical data suggest that my early disappearance to catch a concert did not escape my supervisor’s attention 📊👀
#AcademicSky #PhDSky #Psychology #VLM
However, the empirical data suggest that my early disappearance to catch a concert did not escape my supervisor’s attention 📊👀
#AcademicSky #PhDSky #Psychology #VLM
November 2, 2026 at 11:01 AM
Presenting my VLM research at the Aberdeen Uni Interdisciplinary Symposium was a success!
However, the empirical data suggest that my early disappearance to catch a concert did not escape my supervisor’s attention 📊👀
#AcademicSky #PhDSky #Psychology #VLM
However, the empirical data suggest that my early disappearance to catch a concert did not escape my supervisor’s attention 📊👀
#AcademicSky #PhDSky #Psychology #VLM
New paper: Look Only When Needed: Certificate-Guided Visual Routing for Efficient Diffusion VLMs
TLDR: LOWN cuts diffusion VLM compute by 65% without training by skipping image processing when visual data isn't needed.
https://huggingface.co/papers/2610.36671
#AI #MachineLearning
TLDR: LOWN cuts diffusion VLM compute by 65% without training by skipping image processing when visual data isn't needed.
https://huggingface.co/papers/2610.36671
#AI #MachineLearning
November 1, 2026 at 12:06 AM
New paper: Look Only When Needed: Certificate-Guided Visual Routing for Efficient Diffusion VLMs
TLDR: LOWN cuts diffusion VLM compute by 65% without training by skipping image processing when visual data isn't needed.
https://huggingface.co/papers/2610.36671
#AI #MachineLearning
TLDR: LOWN cuts diffusion VLM compute by 65% without training by skipping image processing when visual data isn't needed.
https://huggingface.co/papers/2610.36671
#AI #MachineLearning
PhD on VLM-Based Planning for Autonomous Mini-Vehicles
• Location: Germany
• Deadline: 2026-11-15
• Host: Technical University of Munich
mlscientist.com/phd-vlm-base...
• Location: Germany
• Deadline: 2026-11-15
• Host: Technical University of Munich
mlscientist.com/phd-vlm-base...
October 31, 2026 at 6:08 PM
PhD on VLM-Based Planning for Autonomous Mini-Vehicles
• Location: Germany
• Deadline: 2026-11-15
• Host: Technical University of Munich
mlscientist.com/phd-vlm-base...
• Location: Germany
• Deadline: 2026-11-15
• Host: Technical University of Munich
mlscientist.com/phd-vlm-base...
拡散VLMの論文 LOWN を、LLaDA-V の実際の推論記録で遊べる3Dゲームにした | Ziyu #zenn
https://zenn.dev/zguo/articles/c86799626b7ece
https://zenn.dev/zguo/articles/c86799626b7ece
拡散VLMの論文 LOWN を、LLaDA-V の実際の推論記録で遊べる3Dゲームにした
zenn.dev
November 1, 2026 at 4:31 PM
拡散VLMの論文 LOWN を、LLaDA-V の実際の推論記録で遊べる3Dゲームにした | Ziyu #zenn
https://zenn.dev/zguo/articles/c86799626b7ece
https://zenn.dev/zguo/articles/c86799626b7ece
ModelScope:
DML: www.modelscope.ai/models/Groun...
VLM: www.modelscope.ai/models/Groun...
GitHub: github.com/groundingpi/...
DML: www.modelscope.ai/models/Groun...
VLM: www.modelscope.ai/models/Groun...
GitHub: github.com/groundingpi/...
October 31, 2026 at 10:05 AM
ModelScope:
DML: www.modelscope.ai/models/Groun...
VLM: www.modelscope.ai/models/Groun...
GitHub: github.com/groundingpi/...
DML: www.modelscope.ai/models/Groun...
VLM: www.modelscope.ai/models/Groun...
GitHub: github.com/groundingpi/...
cool result for child learning: a stronger vision encoder (internet-pretrained) made a VLM far better at learning words from children's video
takeaway: visual pretraining pays off beyond vision. Better self-supervised visual learning could unlock "developmentally-efficient" multimodal learning...
takeaway: visual pretraining pays off beyond vision. Better self-supervised visual learning could unlock "developmentally-efficient" multimodal learning...
How do children learn the meanings of their first words? One idea is cross-situational learning: tracking which words go with which things across many situations. In a new preprint, we test this idea at scale on video from children's own perspective 🧵 arxiv.org/abs/2610.31842
October 28, 2026 at 6:09 PM
cool result for child learning: a stronger vision encoder (internet-pretrained) made a VLM far better at learning words from children's video
takeaway: visual pretraining pays off beyond vision. Better self-supervised visual learning could unlock "developmentally-efficient" multimodal learning...
takeaway: visual pretraining pays off beyond vision. Better self-supervised visual learning could unlock "developmentally-efficient" multimodal learning...
1142 détenus pour 587 places à VLM, qui approche du taux d'occupation de 200%...
October 31, 2026 at 10:54 AM
1142 détenus pour 587 places à VLM, qui approche du taux d'occupation de 200%...
₹2499/-https://royalanarkali.com/product/purple-fox-georgette-embroidery-work-salwar-suit/
Product Code :- VLM 226 Purple
Product Code :- VLM 226 Purple
October 27, 2026 at 11:39 PM
₹2499/-https://royalanarkali.com/product/purple-fox-georgette-embroidery-work-salwar-suit/
Product Code :- VLM 226 Purple
Product Code :- VLM 226 Purple
I've been doing this a lot. I built a harness for VLM tuning on small parameter (~8B) locally hosted models and it's crazy how much improvement we've seen in accuracy with each iteretion. Overall result saves dozens of man-hours/week on menial tasks and the whole thing runs on a few 1080Tis lmao
October 30, 2026 at 10:22 PM
I've been doing this a lot. I built a harness for VLM tuning on small parameter (~8B) locally hosted models and it's crazy how much improvement we've seen in accuracy with each iteretion. Overall result saves dozens of man-hours/week on menial tasks and the whole thing runs on a few 1080Tis lmao
Keeping the VLM backbone frozen and storing the decision experience separately — that's the layer of SpatialBrain worth paying attention to.
October 31, 2026 at 12:00 AM
Keeping the VLM backbone frozen and storing the decision experience separately — that's the layer of SpatialBrain worth paying attention to.
October 31, 2026 at 1:25 AM
VDE-VLM-ACCESS Just Updated: A GitHub Project With No Official Releases Yet + Video
No Official Release Published The GitHub repository VDE-VLM-ACCESS, maintained by i2qq, currently has no published releases. The repository's Releases page displays GitHub's standard message indicating that no…
No Official Release Published The GitHub repository VDE-VLM-ACCESS, maintained by i2qq, currently has no published releases. The repository's Releases page displays GitHub's standard message indicating that no…
VDE-VLM-ACCESS Just Updated: A GitHub Project With No Official Releases Yet + Video
No Official Release Published The GitHub repository VDE-VLM-ACCESS, maintained by i2qq, currently has no published releases. The repository's Releases page displays GitHub's standard message indicating that no releases are available. This means there are currently no official versioned release packages, downloadable binaries, or published release notes available through GitHub's release system. What This Means for Developers The absence of published releases does not necessarily mean the project is inactive or abandoned.
undercodenews.com
October 30, 2026 at 3:57 PM
VDE-VLM-ACCESS Just Updated: A GitHub Project With No Official Releases Yet + Video
No Official Release Published The GitHub repository VDE-VLM-ACCESS, maintained by i2qq, currently has no published releases. The repository's Releases page displays GitHub's standard message indicating that no…
No Official Release Published The GitHub repository VDE-VLM-ACCESS, maintained by i2qq, currently has no published releases. The repository's Releases page displays GitHub's standard message indicating that no…
Our main contributions:
1) PanoCaps, a human-annotated benchmark with dense captions and near-complete pixel coverage
2) gPQ, a metric that jointly scores text and mask agreement
3) PANORAMA, a VLM that grounds phrases by selecting from phrase-conditioned mask proposals
1) PanoCaps, a human-annotated benchmark with dense captions and near-complete pixel coverage
2) gPQ, a metric that jointly scores text and mask agreement
3) PANORAMA, a VLM that grounds phrases by selecting from phrase-conditioned mask proposals
November 2, 2026 at 1:46 PM
Our main contributions:
1) PanoCaps, a human-annotated benchmark with dense captions and near-complete pixel coverage
2) gPQ, a metric that jointly scores text and mask agreement
3) PANORAMA, a VLM that grounds phrases by selecting from phrase-conditioned mask proposals
1) PanoCaps, a human-annotated benchmark with dense captions and near-complete pixel coverage
2) gPQ, a metric that jointly scores text and mask agreement
3) PANORAMA, a VLM that grounds phrases by selecting from phrase-conditioned mask proposals
Open Access UCL Research: From Masks to Pixels and Meaning: A New Taxonomy, Benchmark, and Metrics for VLM Image Tampering
discovery.ucl.ac.uk/id/eprint/10...
discovery.ucl.ac.uk/id/eprint/10...
From Masks to Pixels and Meaning: A New Taxonomy, Benchmark, and Metrics for VLM Image Tampering
-
UCL Discovery
UCL Discovery is UCL's open access repository, showcasing and providing access to UCL research outputs from all UCL disciplines.
discovery.ucl.ac.uk
October 30, 2026 at 11:15 AM
Open Access UCL Research: From Masks to Pixels and Meaning: A New Taxonomy, Benchmark, and Metrics for VLM Image Tampering
discovery.ucl.ac.uk/id/eprint/10...
discovery.ucl.ac.uk/id/eprint/10...
₹2499/-https://royalanarkali.com/product/pink-fox-georgette-embroidery-work-salwar-suit/
Product Code :- VLM 226 Pink
Product Code :- VLM 226 Pink
October 28, 2026 at 11:39 AM
₹2499/-https://royalanarkali.com/product/pink-fox-georgette-embroidery-work-salwar-suit/
Product Code :- VLM 226 Pink
Product Code :- VLM 226 Pink
₹2499/-https://royalanarkali.com/product/green-fox-georgette-embroidery-work-salwar-suit/
Product Code :- VLM 226 Green
Product Code :- VLM 226 Green
October 28, 2026 at 2:40 AM
₹2499/-https://royalanarkali.com/product/green-fox-georgette-embroidery-work-salwar-suit/
Product Code :- VLM 226 Green
Product Code :- VLM 226 Green
On a Mac there's one catch for now: mlx-vlm support is still being upstreamed, so that build needs mlx-vlm installed from a git branch.
October 27, 2026 at 5:20 PM
On a Mac there's one catch for now: mlx-vlm support is still being upstreamed, so that build needs mlx-vlm installed from a git branch.
CPU 파서 vs 최신 VLM, 문서 파싱 성능 비교: PyMuPDF와 GPT-6 Astra
RAG나 문서 분석 파이프라인을 만들다 보면 PDF 파싱 단계에서 고민이 생깁니다. 최근에는 VLM에 PDF를 통째로 넣는 방식이 많이 쓰이지만, 페이지당 API 비용과 처리 시간이 만만치 않습니다. 반대로 CPU 기반 파서는 빠르고 싸지만, "정확도는 떨어지지 않나?"라는 의문이 따라붙습니다.
그래서 공개 벤치마크에서 CPU 파서인 PyMuPDF와 최신 VLM인 GPT-6 Astra를 같은 문서로 비교해 봤습니다.
실험 설정
• …
RAG나 문서 분석 파이프라인을 만들다 보면 PDF 파싱 단계에서 고민이 생깁니다. 최근에는 VLM에 PDF를 통째로 넣는 방식이 많이 쓰이지만, 페이지당 API 비용과 처리 시간이 만만치 않습니다. 반대로 CPU 기반 파서는 빠르고 싸지만, "정확도는 떨어지지 않나?"라는 의문이 따라붙습니다.
그래서 공개 벤치마크에서 CPU 파서인 PyMuPDF와 최신 VLM인 GPT-6 Astra를 같은 문서로 비교해 봤습니다.
실험 설정
• …
CPU 파서 vs 최신 VLM, 문서 파싱 성능 비교: PyMuPDF와 GPT-6 Astra
RAG나 문서 분석 파이프라인을 만들다 보면 PDF 파싱 단계에서 고민이 생깁니다. 최근에는 VLM에 PDF를 통째로 넣는 방식이 많이 쓰이지만, 페이지당 API 비용과 처리 시간이 만만치 않습니다. 반대로 CPU 기반 파서는 빠르고 싸지만, "정확도는 떨어지지 않나?"라는 의문이 따라붙습니다.
그래서 공개 벤치마크에서 CPU 파서인 PyMuPDF와 최신 VLM인 GPT-6 Astra를 같은 문서로 비교해 봤습니다.
실험 설정
• 벤치마크: ParseBench (LlamaIndex가 공개한 문서 파싱 벤치마크)
• 비교 대상: GPT-6 Astra는 ParseBench 리더보드와 동일한 설정으로 실행했고, 같은 문서에 대한 PyMuPDF 결과와 비교했습니다.
• 실행 환경: PyMuPDF는 GPU 없이 CPU만 사용했고, GPT-6 Astra는 외부 API로 호출했습니다.
• 속도·비용 측정: 30페이...
discuss.pytorch.kr
October 29, 2026 at 7:55 AM
CPU 파서 vs 최신 VLM, 문서 파싱 성능 비교: PyMuPDF와 GPT-6 Astra
RAG나 문서 분석 파이프라인을 만들다 보면 PDF 파싱 단계에서 고민이 생깁니다. 최근에는 VLM에 PDF를 통째로 넣는 방식이 많이 쓰이지만, 페이지당 API 비용과 처리 시간이 만만치 않습니다. 반대로 CPU 기반 파서는 빠르고 싸지만, "정확도는 떨어지지 않나?"라는 의문이 따라붙습니다.
그래서 공개 벤치마크에서 CPU 파서인 PyMuPDF와 최신 VLM인 GPT-6 Astra를 같은 문서로 비교해 봤습니다.
실험 설정
• …
RAG나 문서 분석 파이프라인을 만들다 보면 PDF 파싱 단계에서 고민이 생깁니다. 최근에는 VLM에 PDF를 통째로 넣는 방식이 많이 쓰이지만, 페이지당 API 비용과 처리 시간이 만만치 않습니다. 반대로 CPU 기반 파서는 빠르고 싸지만, "정확도는 떨어지지 않나?"라는 의문이 따라붙습니다.
그래서 공개 벤치마크에서 CPU 파서인 PyMuPDF와 최신 VLM인 GPT-6 Astra를 같은 문서로 비교해 봤습니다.
실험 설정
• …
CPU 파서 vs 최신 VLM, 문서 파싱 성능 비교: PyMuPDF와 GPT-6 Astra
RAG나 문서 분석 파이프라인을 만들다 보면 PDF 파싱 단계에서 고민이 생깁니다. 최근에는 VLM에 PDF를 통째로 넣는 방식이 많이 쓰이지만, 페이지당 API 비용과 처리 시간이 만만치 않습니다. 반대로 CPU 기반 파서는 빠르고 싸지만, "정확도는 떨어지지 않나?"라는 의문이 따라붙습니다.
그래서 공개 벤치마크에서 CPU 파서인 PyMuPDF와 최신 VLM인 GPT-6 Astra를 같은 문서로 비교해 봤습니다.
실험 설정
• …
RAG나 문서 분석 파이프라인을 만들다 보면 PDF 파싱 단계에서 고민이 생깁니다. 최근에는 VLM에 PDF를 통째로 넣는 방식이 많이 쓰이지만, 페이지당 API 비용과 처리 시간이 만만치 않습니다. 반대로 CPU 기반 파서는 빠르고 싸지만, "정확도는 떨어지지 않나?"라는 의문이 따라붙습니다.
그래서 공개 벤치마크에서 CPU 파서인 PyMuPDF와 최신 VLM인 GPT-6 Astra를 같은 문서로 비교해 봤습니다.
실험 설정
• …
CPU 파서 vs 최신 VLM, 문서 파싱 성능 비교: PyMuPDF와 GPT-6 Astra
RAG나 문서 분석 파이프라인을 만들다 보면 PDF 파싱 단계에서 고민이 생깁니다. 최근에는 VLM에 PDF를 통째로 넣는 방식이 많이 쓰이지만, 페이지당 API 비용과 처리 시간이 만만치 않습니다. 반대로 CPU 기반 파서는 빠르고 싸지만, "정확도는 떨어지지 않나?"라는 의문이 따라붙습니다.
그래서 공개 벤치마크에서 CPU 파서인 PyMuPDF와 최신 VLM인 GPT-6 Astra를 같은 문서로 비교해 봤습니다.
실험 설정
• 벤치마크: ParseBench (LlamaIndex가 공개한 문서 파싱 벤치마크)
• 비교 대상: GPT-6 Astra는 ParseBench 리더보드와 동일한 설정으로 실행했고, 같은 문서에 대한 PyMuPDF 결과와 비교했습니다.
• 실행 환경: PyMuPDF는 GPU 없이 CPU만 사용했고, GPT-6 Astra는 외부 API로 호출했습니다.
• 속도·비용 측정: 30페이...
discuss.pytorch.kr
October 29, 2026 at 7:46 AM
CPU 파서 vs 최신 VLM, 문서 파싱 성능 비교: PyMuPDF와 GPT-6 Astra
RAG나 문서 분석 파이프라인을 만들다 보면 PDF 파싱 단계에서 고민이 생깁니다. 최근에는 VLM에 PDF를 통째로 넣는 방식이 많이 쓰이지만, 페이지당 API 비용과 처리 시간이 만만치 않습니다. 반대로 CPU 기반 파서는 빠르고 싸지만, "정확도는 떨어지지 않나?"라는 의문이 따라붙습니다.
그래서 공개 벤치마크에서 CPU 파서인 PyMuPDF와 최신 VLM인 GPT-6 Astra를 같은 문서로 비교해 봤습니다.
실험 설정
• …
RAG나 문서 분석 파이프라인을 만들다 보면 PDF 파싱 단계에서 고민이 생깁니다. 최근에는 VLM에 PDF를 통째로 넣는 방식이 많이 쓰이지만, 페이지당 API 비용과 처리 시간이 만만치 않습니다. 반대로 CPU 기반 파서는 빠르고 싸지만, "정확도는 떨어지지 않나?"라는 의문이 따라붙습니다.
그래서 공개 벤치마크에서 CPU 파서인 PyMuPDF와 최신 VLM인 GPT-6 Astra를 같은 문서로 비교해 봤습니다.
실험 설정
• …
₹2499/-https://royalanarkali.com/product/firozi-crunchy-silk-lace-embroidery-work-gown/
Product Code :- VLM 297 Firozi
Product Code :- VLM 297 Firozi
October 27, 2026 at 5:40 PM
₹2499/-https://royalanarkali.com/product/firozi-crunchy-silk-lace-embroidery-work-gown/
Product Code :- VLM 297 Firozi
Product Code :- VLM 297 Firozi
₹2499/-https://royalanarkali.com/product/cream-crunchy-silk-lace-embroidery-work-gown/
Product Code :- VLM 297 Cream
Product Code :- VLM 297 Cream
October 27, 2026 at 2:40 PM
₹2499/-https://royalanarkali.com/product/cream-crunchy-silk-lace-embroidery-work-gown/
Product Code :- VLM 297 Cream
Product Code :- VLM 297 Cream
PANORAMA: for each phrase the VLM emits a [SEG] token, projected into a concept vector that conditions SAM 3 to propose candidate masks. A learned scorer then picks the ones matching the phrase. This keeps SAM 3's mask quality and enhances it by selecting the best of its proposals for each phrase.
November 2, 2026 at 1:46 PM
PANORAMA: for each phrase the VLM emits a [SEG] token, projected into a concept vector that conditions SAM 3 to propose candidate masks. A learned scorer then picks the ones matching the phrase. This keeps SAM 3's mask quality and enhances it by selecting the best of its proposals for each phrase.
Huan Yang, Shiqi Jiang, Yu Mao, Chuanyue Xiong, Deyu Zhang: How Far Are We From On-Device VLM Inference? A Measurement Study on Edge NPUs https://arxiv.org/abs/2610.40039 https://arxiv.org/pdf/2610.40039 https://arxiv.org/html/2610.40039
October 30, 2026 at 6:43 AM
Huan Yang, Shiqi Jiang, Yu Mao, Chuanyue Xiong, Deyu Zhang: How Far Are We From On-Device VLM Inference? A Measurement Study on Edge NPUs https://arxiv.org/abs/2610.40039 https://arxiv.org/pdf/2610.40039 https://arxiv.org/html/2610.40039
Canva lança o Brush, uma ferramenta de IA que reorganiza projetos completos com inteligência artificial.
#custo #desempenho #design #ferramenta #modelo
#custo #desempenho #design #ferramenta #modelo
Canva lança o Brush, uma ferramenta de IA que reorganiza projetos completos
A plataforma de design Canva acaba de apresentar o Canva Brush, um modelo de visão e linguagem (VLM) concebido especificamente para modificar imagens e estruturas gráficas completas através de simples instruções escritas. Ao contrário dos geradores tradicionais que produzem imagens estáticas achatad
tugatech.com.pt
October 30, 2026 at 12:42 PM
Canva lança o Brush, uma ferramenta de IA que reorganiza projetos completos com inteligência artificial.
#custo #desempenho #design #ferramenta #modelo
#custo #desempenho #design #ferramenta #modelo
