Ilyass Moummad
@ilyassmoummad.bsky.social
Postdoctoral Researcher @ Inria Montpellier (IROKO, Pl@ntNet)
SSL for plant images
Interested in Computer Vision, Natural Language Processing, Machine Listening, and Biodiversity Monitoring
Website: ilyassmoummad.github.io
SSL for plant images
Interested in Computer Vision, Natural Language Processing, Machine Listening, and Biodiversity Monitoring
Website: ilyassmoummad.github.io
[10/10] Wrap-up 🎯
🔹 Unified supervised + unsupervised hashing
🔹 Flexible: works via probing or LoRA
🔹 SOTA hashing in minutes on a single GPU
📄 Paper: arxiv.org/abs/2510.27584
💻 Code: github.com/ilyassmoumma...
Shoutout to my wonderful co-authors Kawtar, Hervé, and Alexis.
🔹 Unified supervised + unsupervised hashing
🔹 Flexible: works via probing or LoRA
🔹 SOTA hashing in minutes on a single GPU
📄 Paper: arxiv.org/abs/2510.27584
💻 Code: github.com/ilyassmoumma...
Shoutout to my wonderful co-authors Kawtar, Hervé, and Alexis.
November 3, 2025 at 2:31 PM
[10/10] Wrap-up 🎯
🔹 Unified supervised + unsupervised hashing
🔹 Flexible: works via probing or LoRA
🔹 SOTA hashing in minutes on a single GPU
📄 Paper: arxiv.org/abs/2510.27584
💻 Code: github.com/ilyassmoumma...
Shoutout to my wonderful co-authors Kawtar, Hervé, and Alexis.
🔹 Unified supervised + unsupervised hashing
🔹 Flexible: works via probing or LoRA
🔹 SOTA hashing in minutes on a single GPU
📄 Paper: arxiv.org/abs/2510.27584
💻 Code: github.com/ilyassmoumma...
Shoutout to my wonderful co-authors Kawtar, Hervé, and Alexis.
[9/10] Strong generalization 🌍
CroVCA produces compact codes that transfer efficiently:
✅ Single HashCoder trained on ImageNet-1k works on downstream datasets without retraining (More experiments and ablations in the paper)
CroVCA produces compact codes that transfer efficiently:
✅ Single HashCoder trained on ImageNet-1k works on downstream datasets without retraining (More experiments and ablations in the paper)
November 3, 2025 at 2:31 PM
[9/10] Strong generalization 🌍
CroVCA produces compact codes that transfer efficiently:
✅ Single HashCoder trained on ImageNet-1k works on downstream datasets without retraining (More experiments and ablations in the paper)
CroVCA produces compact codes that transfer efficiently:
✅ Single HashCoder trained on ImageNet-1k works on downstream datasets without retraining (More experiments and ablations in the paper)
[8/10] Semantically consistent retrieval 🔍
CroVCA retrieves correct classes even for fine-grained or ambiguous queries (e.g., indigo bird, grey langur).
✅ Outperforms Hashing-Baseline
✅ Works with only 16 bits and without supervision
CroVCA retrieves correct classes even for fine-grained or ambiguous queries (e.g., indigo bird, grey langur).
✅ Outperforms Hashing-Baseline
✅ Works with only 16 bits and without supervision
November 3, 2025 at 2:31 PM
[8/10] Semantically consistent retrieval 🔍
CroVCA retrieves correct classes even for fine-grained or ambiguous queries (e.g., indigo bird, grey langur).
✅ Outperforms Hashing-Baseline
✅ Works with only 16 bits and without supervision
CroVCA retrieves correct classes even for fine-grained or ambiguous queries (e.g., indigo bird, grey langur).
✅ Outperforms Hashing-Baseline
✅ Works with only 16 bits and without supervision
[7/10] Compact yet meaningful codes 💾
Even with just 16 bits, CroVCA preserves class structure.
t-SNE on CIFAR-10 shows clear, separable clusters — almost identical to the original 768-dim embeddings.
Even with just 16 bits, CroVCA preserves class structure.
t-SNE on CIFAR-10 shows clear, separable clusters — almost identical to the original 768-dim embeddings.
November 3, 2025 at 2:31 PM
[7/10] Compact yet meaningful codes 💾
Even with just 16 bits, CroVCA preserves class structure.
t-SNE on CIFAR-10 shows clear, separable clusters — almost identical to the original 768-dim embeddings.
Even with just 16 bits, CroVCA preserves class structure.
t-SNE on CIFAR-10 shows clear, separable clusters — almost identical to the original 768-dim embeddings.
[6/10] Strong performance across encoders 💪
Tested on multiple vision encoders (SimDINOv2, DINOv2, DFN…), CroVCA achieves SOTA unsupervised hashing:
Tested on multiple vision encoders (SimDINOv2, DINOv2, DFN…), CroVCA achieves SOTA unsupervised hashing:
November 3, 2025 at 2:30 PM
[6/10] Strong performance across encoders 💪
Tested on multiple vision encoders (SimDINOv2, DINOv2, DFN…), CroVCA achieves SOTA unsupervised hashing:
Tested on multiple vision encoders (SimDINOv2, DINOv2, DFN…), CroVCA achieves SOTA unsupervised hashing:
[5/10] Fast convergence 🚀
CroVCA trains in just ~5 epochs:
✅ COCO (unsupervised) <2 min
✅ ImageNet100 (supervised) ~3 min
✅ Single GPU
Despite simplicity, it achieves state-of-the-art retrieval performance.
CroVCA trains in just ~5 epochs:
✅ COCO (unsupervised) <2 min
✅ ImageNet100 (supervised) ~3 min
✅ Single GPU
Despite simplicity, it achieves state-of-the-art retrieval performance.
November 3, 2025 at 2:30 PM
[5/10] Fast convergence 🚀
CroVCA trains in just ~5 epochs:
✅ COCO (unsupervised) <2 min
✅ ImageNet100 (supervised) ~3 min
✅ Single GPU
Despite simplicity, it achieves state-of-the-art retrieval performance.
CroVCA trains in just ~5 epochs:
✅ COCO (unsupervised) <2 min
✅ ImageNet100 (supervised) ~3 min
✅ Single GPU
Despite simplicity, it achieves state-of-the-art retrieval performance.
[4/10] HashCoder 🛠️
A lightweight MLP with final BatchNorm for balanced bits (inspired by OrthoHash). Can be used as:
🔹 Probe on frozen features
🔹 LoRA-based fine-tuning for efficient encoder adaptation
A lightweight MLP with final BatchNorm for balanced bits (inspired by OrthoHash). Can be used as:
🔹 Probe on frozen features
🔹 LoRA-based fine-tuning for efficient encoder adaptation
November 3, 2025 at 2:30 PM
[4/10] HashCoder 🛠️
A lightweight MLP with final BatchNorm for balanced bits (inspired by OrthoHash). Can be used as:
🔹 Probe on frozen features
🔹 LoRA-based fine-tuning for efficient encoder adaptation
A lightweight MLP with final BatchNorm for balanced bits (inspired by OrthoHash). Can be used as:
🔹 Probe on frozen features
🔹 LoRA-based fine-tuning for efficient encoder adaptation
[3/10] Unifying hashing 🔄
Can supervised + unsupervised hashing be done in one framework?
CroVCA aligns binary codes across semantically consistent views:
Augmentations → unsupervised
Class-consistent samples → supervised
🧩 One BCE loss + coding-rate regularizer
Can supervised + unsupervised hashing be done in one framework?
CroVCA aligns binary codes across semantically consistent views:
Augmentations → unsupervised
Class-consistent samples → supervised
🧩 One BCE loss + coding-rate regularizer
November 3, 2025 at 2:30 PM
[3/10] Unifying hashing 🔄
Can supervised + unsupervised hashing be done in one framework?
CroVCA aligns binary codes across semantically consistent views:
Augmentations → unsupervised
Class-consistent samples → supervised
🧩 One BCE loss + coding-rate regularizer
Can supervised + unsupervised hashing be done in one framework?
CroVCA aligns binary codes across semantically consistent views:
Augmentations → unsupervised
Class-consistent samples → supervised
🧩 One BCE loss + coding-rate regularizer
[2/10] The challenge ⚡
Foundation models (DINOv3, DFN, SWAG…) produce rich embeddings, but similarity search in high-dimensional spaces is expensive.
Hashing provides fast Hamming-distance search, yet most deep hashing methods are complex, slow, and tied to a single paradigm.
Foundation models (DINOv3, DFN, SWAG…) produce rich embeddings, but similarity search in high-dimensional spaces is expensive.
Hashing provides fast Hamming-distance search, yet most deep hashing methods are complex, slow, and tied to a single paradigm.
November 3, 2025 at 2:29 PM
[2/10] The challenge ⚡
Foundation models (DINOv3, DFN, SWAG…) produce rich embeddings, but similarity search in high-dimensional spaces is expensive.
Hashing provides fast Hamming-distance search, yet most deep hashing methods are complex, slow, and tied to a single paradigm.
Foundation models (DINOv3, DFN, SWAG…) produce rich embeddings, but similarity search in high-dimensional spaces is expensive.
Hashing provides fast Hamming-distance search, yet most deep hashing methods are complex, slow, and tied to a single paradigm.
I heard that the Linux client is buggy, I use it on the browser and it's working ok.
September 9, 2025 at 7:03 AM
I heard that the Linux client is buggy, I use it on the browser and it's working ok.
for the curious, the code, slides and the article are on Github: github.com/BastienPasde...
August 29, 2025 at 11:44 AM
for the curious, the code, slides and the article are on Github: github.com/BastienPasde...
love it haha wish I were there to hear Prostitute Disfigurement in an amphitheater
August 29, 2025 at 11:40 AM
love it haha wish I were there to hear Prostitute Disfigurement in an amphitheater
Im interested in the quantum and footnotesize, how much params should they have 😂
August 23, 2025 at 6:31 AM
Im interested in the quantum and footnotesize, how much params should they have 😂
It feels like we can now fit more noise with more model capacity 🤔 (Figure 6), maybe we need newer architectures and/or newer training losses.
August 19, 2025 at 9:36 PM
It feels like we can now fit more noise with more model capacity 🤔 (Figure 6), maybe we need newer architectures and/or newer training losses.
👋 I worked on bioacoustics during my PhD, but I post mostly about AI
July 18, 2025 at 8:56 PM
👋 I worked on bioacoustics during my PhD, but I post mostly about AI
my new addiction today: youtu.be/dSyJqwN36ow
I can't wait to see them this summer in Motocultor Festival
I can't wait to see them this summer in Motocultor Festival
Of Petrichor Weaves Black Noise
YouTube video by Ne Obliviscaris - Topic
youtu.be
June 19, 2025 at 9:54 AM
my new addiction today: youtu.be/dSyJqwN36ow
I can't wait to see them this summer in Motocultor Festival
I can't wait to see them this summer in Motocultor Festival
the best discovery I've had in recent years, I'm addicted to it now as well 😁
June 19, 2025 at 7:12 AM
the best discovery I've had in recent years, I'm addicted to it now as well 😁
Thank you for making this accessible to everyone! I've read some sections, it is very instructive.
June 16, 2025 at 10:08 AM
Thank you for making this accessible to everyone! I've read some sections, it is very instructive.