# EarthJev: Calibrated, Non-Generative Answers to Natural-Language Questions about Sentinel-2 Land Cover

DevFeed: [EarthJev: Calibrated, Non-Generative Answers to Natural-Language Questions about Sentinel-2 Land Cover](<https://devfeed.tech/articles/earthjev-calibrated-non-generative-answers-to-natural-language-questions-about-sentinel-2-land-cover-115268.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.02096>)

Author: Domenico Benfenati, Antonio Maria Rinaldi, Cristian Tommasino

Published: 2026-11-03T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [vlm](<https://devfeed.tech/topics/vlm.md>), [Calibrated uncertainty](<https://devfeed.tech/topics/calibrated-uncertainty.md>), [qwen](<https://devfeed.tech/topics/qwen.md>), [lora](<https://devfeed.tech/topics/lora.md>)

Tags: [2](<https://devfeed.tech/tags/2.md>), [ablation](<https://devfeed.tech/tags/ablation.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [decoding](<https://devfeed.tech/tags/decoding.md>), [faster](<https://devfeed.tech/tags/faster.md>), [fixed](<https://devfeed.tech/tags/fixed.md>), [generate](<https://devfeed.tech/tags/generate.md>), [generative](<https://devfeed.tech/tags/generative.md>), [generative-models](<https://devfeed.tech/tags/generative-models.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [image](<https://devfeed.tech/tags/image.md>), [input](<https://devfeed.tech/tags/input.md>), [language](<https://devfeed.tech/tags/language.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [multimodal](<https://devfeed.tech/tags/multimodal.md>), [multimodal-models](<https://devfeed.tech/tags/multimodal-models.md>), [natural-language](<https://devfeed.tech/tags/natural-language.md>), [questions](<https://devfeed.tech/tags/questions.md>), [remote-sensing](<https://devfeed.tech/tags/remote-sensing.md>), [satellite](<https://devfeed.tech/tags/satellite.md>)

## AI overview

EarthJev answers typed questions about Sentinel-2 land-cover patches by selecting among option-letter logits at one position, avoiding token generation and producing calibrated probabilities. Its 0.8B Qwen3.5-based model takes all 12 spectral bands and is trained with cross-entropy plus Brier score. On 36,028 test questions, it achieves 86.0% accuracy and ECE 0.006. At 19.9 ms per question, it is reported as 4.3-8.7 times faster than the compared multimodal language models and 6.8 times faster than its own generative backbone; accuracy trails the two 2B models by 0.6 and 1.2 points and EarthDial-4B by 3.5 points. Replacing the nine non-RGB bands with their mean reduces accuracy by 4.5 points, while an ablation that trains the full vision encoder reaches 88.4% accuracy at the same cost.

## Source excerpt

arXiv:2611.02096v1 Announce Type: cross Abstract: Vision-language models for remote sensing answer questions about satellite images by generating text: every answer costs a decoding loop, has to be parsed back to an option, and comes without a usable probability. We present EarthJev, a 0.8B-parameter model that answers typed questions about multispectral Sentinel-2 patches without generating any token. All 12 bands enter the vision encoder of Qwen3.5-0.8B through a widened input layer initialised from the RGB kernels, and, following the open OneJev recipe, the answer is read from the logits of the option letters at a single position and trained with cross-entropy plus Brier score. Under a pre-specified analysis plan, we compare it with the same backbone fine-tuned to generate the answer and with three multimodal language models fine-tuned with LoRA, all trained on the same 70,890 questions from 5% of BigEarthNet v2.0 and timed in one job on the same GPU; VQA models that classify over a fixed answer vocabulary are reported as a reference for the cost of answering only the questions seen in training. EarthJev answers a question in 19.9 ms, 4.3 to 8.7 times faster than the multimodal language models and 6.8 times faster than its own backbone used generatively; when all the questions about an image are asked together, it answers them in 42.8 ms, 3.7 to 10.1 times faster than the batched generative models. On 36,028 test questions it reaches 86.0% accuracy (three seeds), 0.6 and 1.2 points below the two 2B multimodal models and 3.5 points below the remote-sensing EarthDial-4B, with answers that are valid by construction and calibrated probabilities (ECE 0.006). Replacing the nine non-RGB bands with their mean costs 4.5 points, so the model uses the multispectral input, and in an ablation training the whole vision encoder raises accuracy to 88.4% at the same cost. Code and model: https://github.com/CosmoIknosLab/EarthJev