Research map / Vision-language and multimodal / Human and embodied modeling
KerasCV and KerasNLP: Multi-framework Models
Matthew Watson, Divyashree Shivakumar Sreepathihalli, François Chollet, Martin Górner, Kiranbir Sodhia, Ramesh Sampath, Tirth Patel, Haifeng Jin, Neel Kovelamudi, Gabriel Rasskin, Samaneh Saadat, Luke Wood and 5 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| select the fastest Keras 3 backend instead of Keras 2 (TF) | Segment Anything · Segment Anything predict (batch size 7) | time per step (ms): 3,187.09 -> 660.16 | Improved |
| select the fastest Keras 3 backend instead of Keras 2 (TF) | Segment Anything · Segment Anything train (batch size 1) | time per step (ms): 386.93 -> 355.25 | Improved |
| select the fastest Keras 3 backend instead of Keras 2 (TF) | BERT · BERT train (batch size 54) | time per step (ms): 841.84 -> 404.17 | Improved |
| select the fastest Keras 3 backend instead of Keras 2 (TF) | BERT · BERT predict (batch size 531) | time per step (ms): 965.21 -> 865.29 | Improved |
| use PyTorch backend instead of JAX in Keras 3 | Segment Anything · Segment Anything train (batch size 1) | time per step (ms): 361.69 -> 1,388.87 | Hurt |
| use PyTorch backend instead of JAX in Keras 3 | Mistral · Mistral predict (batch size 32) | time per step (ms): 957.25 -> 10932.59 | Hurt |
| use PyTorch backend instead of TF in Keras 3 | Gemma · Gemma predict (batch size 32) | time per step (ms): 1,134.91 -> 7, 952.67 | Hurt |
| use JAX backend instead of TF in Keras 3 | Gemma · Gemma train (batch size 8) | time per step (ms): 232.52 -> 273.67 | Hurt |
| use JAX backend instead of TF in Keras 3 | BERT · BERT predict (batch size 531) | time per step (ms): 962.11 -> 865.29 | Improved |
| use JAX backend instead of TF in Keras 3 | Mistral · Mistral train (batch size 8) | time per step (ms): 185.92 -> 213.22 | Hurt |
Closest papers by meaning
- GluonCV and GluonNLP: Deep Learning in Computer Vision and Natural Language Processing (JMLR 2020 · similarity 0.85)
- ECoFLaP: Efficient Coarse-to-Fine Layer-Wise Pruning for Vision-Language Models (ICLR 2024 · similarity 0.84)
- Dataset Distillation for Pre-Trained Self-Supervised Vision Models (NeurIPS 2025 · similarity 0.83)
- Scaling Laws for Optimal Data Mixtures (NeurIPS 2025 · similarity 0.83)
- MCNC: Manifold-Constrained Reparameterization for Neural Compression (ICLR 2025 · similarity 0.83)
- Domain2Vec: Vectorizing Datasets to Find the Optimal Data Mixture without Training (ICML 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
