--- language: - en license: apache-2.0 library_name: devision pipeline_tag: visual-question-answering base_model: - convaiinnovations/laya - google/siglip2-base-patch16-256 datasets: - HuggingFaceM4/VQAv2 - lmms-lab/GQA - HuggingFaceM4/A-OKVQA - derek-thomas/ScienceQA - HuggingFaceM4/the_cauldron - cambridgeltl/vsr_random - jxu124/objects365 - J1mb0o/e-SNLI-VE - Vision-Flan/vision-flan_191-task_1k - ryokamoi/VisOnlyQA_Train - RyanWW/Super-CLEVR - allenai/pixmo-count - russwest/AitW - osunlp/Multimodal-Mind2Web - qiup0307/EK100-VISOR - HuggingFaceM4/ChartQA - naver-clova-ix/cord-v2 - MMInstruction/ArxivQA - OpenDriveLab/DriveLM - neuralcatcher/hateful_memes tags: - devision - visual-decisions - calibrated-probabilities - jev - system-one - rlcd - cpu model-index: - name: deVision v0.5 results: - task: type: visual-question-answering dataset: name: COCO object presence type: test_exist metrics: - type: accuracy value: 0.9518 - type: ece value: 0.0705 - task: type: visual-question-answering dataset: name: VQAv2 multiple choice type: test_vqa_choice metrics: - type: accuracy value: 0.905 - type: ece value: 0.0265 - task: type: visual-question-answering dataset: name: COCO size type: test_size metrics: - type: accuracy value: 0.8765 - type: ece value: 0.1535 - task: type: visual-question-answering dataset: name: POPE, project filtered set type: bench_pope metrics: - type: accuracy value: 0.8607 - type: ece value: 0.0285 - task: type: visual-question-answering dataset: name: GQA val subset type: test_gqa metrics: - type: accuracy value: 0.8448 - type: ece value: 0.0291 - task: type: visual-question-answering dataset: name: COCO position type: test_position metrics: - type: accuracy value: 0.9412 - type: ece value: 0.0407 - task: type: visual-question-answering dataset: name: VQAv2 yes/no subset type: test_vqa_yesno metrics: - type: accuracy value: 0.721 - type: ece value: 0.0622 - task: type: visual-question-answering dataset: name: COCO relative position type: test_relation metrics: - type: accuracy value: 0.9121 - type: ece value: 0.0281 - task: type: visual-question-answering dataset: name: VSR, project held-out split type: test_vsr metrics: - type: accuracy value: 0.7184 - type: ece value: 0.1022 - task: type: visual-question-answering dataset: name: Visual7W, project held-out split type: test_v7w metrics: - type: accuracy value: 0.777 - type: ece value: 0.0407 - task: type: visual-question-answering dataset: name: CountBench, 0–10 choices type: bench_countbench metrics: - type: accuracy value: 0.5153 - type: ece value: 0.0791 - task: type: visual-question-answering dataset: name: COCO mirrored position type: test_position_flip metrics: - type: accuracy value: 0.9406 - type: ece value: 0.0337 - task: type: visual-question-answering dataset: name: COCO mirrored relative position type: test_relation_flip metrics: - type: accuracy value: 0.9227 - type: ece value: 0.0196 - task: type: visual-question-answering dataset: name: What'sUp type: bench_whatsup metrics: - type: accuracy value: 0.5183 - type: ece value: 0.096 --- # deVision v0.5

deVision

**Image + English questions → structured answers with calibrated probabilities.** deVision pairs a SigLIP2 vision encoder with Laya's ModernBERT decision model. It scores the options of yes/no (`noul`) and multiple-choice (`choice`) questions instead of generating text, follows the Jev answer format, and runs without a GPU. Several questions about one image share a single image encoding. ## Usage ```bash pip install devision ``` ```python import devision model = devision.load("lukbit/devision") # latest release; device defaults to auto result = model.predict( image="photo.jpg", # a path, an http(s) URL, a data URI, base64, bytes or a PIL image state="", # text context; "" when there is none questions={ "has_fork": {"type": "noul", "instructions": "Is there a fork in the image?"}, "room": {"type": "choice", "instructions": "Which room is this?", "criteria": {"kitchen": None, "bathroom": None, "bedroom": None}}, }, ) print(result["answers"]["has_fork"]["noul"]) # probability of yes print(result["answers"]["room"]["probabilities"]) # one probability per option, summing to 1 ``` HTTP server with a browser demo at `http://127.0.0.1:8000`: ```bash devision-serve --checkpoint lukbit/devision --port 8000 curl -s http://127.0.0.1:8000/v1/systemone -H 'Content-Type: application/json' \ -d '{"image": "https://example.com/photo.jpg", "state": "", "questions": {"has_fork": {"type": "noul", "instructions": "Is there a fork in the image?"}}}' ``` Over HTTP, `image` is an http(s) URL, a data URI or base64; the server never reads its own files. Responses follow Jev; `confidence` for `choice` is `(n * p_max - 1) / (n - 1)`. Invalid requests return 422 (`devision.InvalidRequest` in Python). **Good for:** whether something is present and how many (ask counts as a choice), what kind of thing or scene, colours and materials, up/down and left/right in photos (as a choice). **Not for:** comparing two places on a diagram, chart or map; relative-position yes/no questions; small text; other languages or several images. Use the probabilities as thresholds and send uncertain cases to a person or a stronger model, after checking the thresholds on your own data. ## Evaluation Accuracy through `decide` with the fitted temperatures. *Mismatched*: the same questions with every picture swapped for an unrelated one. These are project evaluation subsets, not official leaderboard scores. Although test pictures are excluded from training, these sets have informed successive experiments and release decisions; they are not a fresh independent test. See the release limitations below. | Set | Questions | Accuracy | Mismatched | NLL | ECE | |---|---:|---:|---:|---:|---:| | COCO object presence | 1,120 | 0.952 | 0.505 | 0.189 | 0.070 | | VQAv2 multiple choice | 1,453 | 0.905 | 0.373 | 0.251 | 0.026 | | COCO size | 1,198 | 0.876 | 0.511 | 0.416 | 0.153 | | POPE, project filtered set | 9,000 | 0.861 | 0.518 | 0.361 | 0.029 | | GQA val subset | 986 | 0.845 | 0.558 | 0.346 | 0.029 | | COCO position | 1,276 | 0.941 | 0.500 | 0.182 | 0.041 | | VQAv2 yes/no subset | 1,000 | 0.721 | 0.501 | 0.567 | 0.062 | | COCO relative position | 2,366 | 0.912 | 0.472 | 0.226 | 0.028 | | VSR, project held-out split | 902 | 0.718 | 0.469 | 0.590 | 0.102 | | Visual7W, project held-out split | 1,000 | 0.777 | 0.446 | 0.585 | 0.041 | | CountBench, 0–10 choices | 491 | 0.515 | 0.106 | 1.261 | 0.079 | | COCO mirrored position | 640 | 0.941 | 0.522 | 0.177 | 0.034 | | COCO mirrored relative position | 1,786 | 0.923 | 0.490 | 0.210 | 0.020 | | What'sUp | 820 | 0.518 | 0.233 | 1.174 | 0.096 | Against Laya Vision 201M on the same questions (its published per-question predictions; difference in points, 95% interval from paired resampling by picture): | Set | Questions | Laya Vision | deVision | Difference | |---|---:|---:|---:|---| | VQAv2 yes/no | 4,887 | 0.717 | 0.749 | +3.2 [+1.7, +4.7] | | A-OKVQA | 1,138 | 0.598 | 0.627 | +2.9 [-0.3, +6.2] | | ScienceQA with images | 2,097 | 0.824 | 0.821 | -0.3 [-2.3, +1.5] | | ScienceQA natural science, needs the picture | 323 | 0.700 | 0.601 | -9.9 [-16.4, -3.4] | On the full POPE random / popular / adversarial sets (3,000 questions each) deVision scores **0.894 / 0.867 / 0.821**, against Laya Vision's published **0.836 / 0.819 / 0.777** (aggregate scores only, not paired). CPU latency: P50 301 ms, P95 321 ms (Linux x86_64, 4 threads, FP32, one question per request, warm-up excluded). ## Model | | | |---|---| | Vision | SigLIP2-B/16, 256 × 256 letterboxed input, 64 visual tokens after a 2 × 2 merge and an MLP projector; training policy described below | | Decision | Laya-initialised ModernBERT-large and Laya's decision head; each option is scored at its `[MASK]`, then a temperature-scaled softmax | | Size | About 518M parameters, FP32, 2.07 GB (LoRA merged) | This release continues v0.4 in one pass on one RTX 5070 (bf16), with SigLIP2's top four encoder layers and final LayerNorm fine-tuned at 5e-6 alongside the projector, decision head and ModernBERT LoRA (merged into the released weights). The input remains 256×256 with 64 visual tokens; the architecture is unchanged. **268,583 questions, one pass from v0.4**, aimed at everyday uses v0.4 could not handle: - phone screenshots (Android in the Wild: general and shopping tasks, with the screen state as the question) and web pages (Mind2Web: which marked element to act on), ScreenQA; - first-person everyday scenes (EPIC-KITCHENS VISOR: what is being held, touched or done) and driving scenes (DriveLM, front camera); - charts (ChartQA, distractors from the same chart), paper figures (ArxivQA), diagrams and science (AI2D, TQA, ScienceQA, IconQA, CLEVR); - receipts and forms (CORD, FUNSD: which value a field holds), image–caption matching with hard negatives (COCO captions), meme moderation (Hateful Memes train), relative depth (COCO indoor photos labelled by a monocular depth model, kept only where the depth gap is large); - replay of v0.4's natural-photo abilities (VQAv2, GQA, presence, size, position, relations, counting, Visual7W, VSR, A-OKVQA), whose targets were set to 0.5 × gold + 0.5 × v0.4's own answer so that the old abilities stay where they were. Every picture of every evaluation set was excluded by id and image hash, including the pictures of the external benchmarks we report against (145 training rows were removed by this filter). Temperatures were fitted at the end of training on a 1,362-question development set (charts, yes/no, presence, VQA and GQA choice), by question type and option-count bucket: | Bucket | Temperature | |---|---:| | Choice, 2 options | 1.1246 | | Choice, 3–5 options | 1.1677 | | Noul (yes / no) | 0.9890 | | Choice, any other count | 1.1601 | Against v0.4 on the same questions (95% intervals from resampling pictures; full table in accept_vs_v0.4.md): | set | v0.4 | v0.5 | difference | |---|---|---|---| | phone shopping screens (AitW, state questions) | 21.4% | 68.5% | +47.0 [+42.8, +51.2] | | phone / web screens (AitW, Mind2Web) | 38.6% | 78.2% | +39.7 [+36.8, +42.7] | | phone screen state | 25.3% | 52.4% | +27.1 [+24.0, +30.1] | | first-person everyday scenes (EPIC-KITCHENS) | 34.7% | 71.7% | +37.0 [+35.4, +38.7] | | charts (ChartQA) | 30.7% | 58.9% | +28.2 [+25.7, +30.9] | | CV-Bench | 55.1% | 60.8% | +5.7 [+4.0, +7.3] | | What'sUp | 46.7% | 51.8% | +5.1 [+2.7, +7.6] | | POPE | 86.5% | 86.1% | −0.5 [−1.1, +0.1] | On held-out pictures of the new training sources: paper figures 32.0% → 51.5%, receipts / forms 39.3% → 53.4%, driving scenes 43.5% → 75.7%, caption matching 85.9% → 94.4%, memes 50.8% → 60.9%, relative depth 50.4% → 55.4% (all intervals above 0). No other evaluation set (MMBench, MME, NaturalBench, RealWorldQA, CountBench, EgoThink, CoGenT, the laya-vision sets, our presence / position / relation / size / VSR / Visual7W / VQA tests) is significantly worse. ## Limitations - **Scope.** English, one image, yes/no and choice questions only; `score` is not implemented. At 256×256, small text (dense documents, infographics, desktop web pages) is hard to read. - **POPE.** About 0.5 points below v0.4 on POPE (86.1% vs 86.5%, interval [−1.1, +0.1]) (object-presence hallucination); the owner accepted this small regression for the new abilities. - **Web pages.** Our held-out phone and web screens improved strongly, but choosing the marked element to act on in dense desktop web pages stays near chance. - **Counting and science** did not improve over v0.4 (CountBench 51.5% vs 50.1%, COCO counting 35.0% vs 34.5%, ScienceQA 82.1% vs 82.1%, none significant). - **Calibration.** NLL is lower than or equal to v0.4's on 27 of 30 sets (strongly on the new uses), but the model is more over-confident on NaturalBench (NLL +0.029, significant); ECE is also higher on RealWorldQA, MMBench, NaturalBench and a few others (by 0.003–0.029). Refitting the temperatures on the development sets alone narrows but does not remove this. - **Labels.** The relative-depth questions use a depth model's predictions, not measured depth; the memes source has its own research licence. Several sources are non-commercial; see the dataset list. - **Test-set reuse.** Test sets informed several decisions in this cycle. This is one training seed. - **Release decision and original goals.** The owner approved r24 as v0.5 on 2026-10-30 with the calibration trade-offs above. This release does not establish official ImageJevBench v0.3.0 admission, sub-1B leadership, or improvement across its five official domains. The five project-use evaluations above are a different protocol; no official rank is claimed. ## Licence Weights and code: Apache-2.0, like the three base models ([Laya](https://huggingface.co/convaiinnovations/laya), [SigLIP2](https://huggingface.co/google/siglip2-base-patch16-256), [ModernBERT](https://huggingface.co/answerdotai/ModernBERT-large)). The training datasets (listed in this card's metadata) have their own terms, some non-commercial (e.g. ScienceQA, CC BY-NC-SA 4.0); check them against your use. ## Links [Code and training records](https://github.com/byebyebruce/devision) (this release: tag `model-v0.5`) · [evaluation details](evaluation/results.md) · [provenance](provenance.json) · [Laya Vision](https://github.com/r33drichards/laya-vision)