ZacharyZhang-NY
03ca35763f
A screenshot with no OCR-able words used to be a memory nothing could reach. It now has a vector, and a question can be asked with it. - picture space: Xenova/chinese-clip-vit-base-patch16 fp16 ONNX, 361MB, sha256 0fad978faa89b85f…, CPU-runnable on the onnxruntime the repo already ships. Text 76ms, image 29ms/image at batch 8. The ONNX contract was measured, not assumed: it is the fused contrastive graph, so the unused tower is fed a placeholder row, and the tokenizer must be BertTokenizer — a CLIP tokenizer emits ids past the embedding table. - visual_index.py keeps picture vectors in their own family beside the text ones, keyed by atom id, and refuses to answer when the map and the array disagree. retrieval.py stays under 500 lines. - a search can be asked with a picture, and when the words come up thin the picture ranking joins the fusion — but only then, so a text query never pays for the encoder. - visual tokens are counted at 32x32 patches (migration v7): 21,325 image atoms went from 5,331 to 34,618 tokens, and the library from 29.9M to 84.2M. The 100M capsule fits this 50,278-file corpus and nothing much larger — that is what the claim actually means. - the vision-language describer is written and gated: with no snapshot an atom records "deferred: no vision-language model snapshot"; a broken model becomes a note on the atom instead of a dead ingest. - forget now voids the picture index and the copy this process holds, so a forgotten picture stops answering from RAM. Measured, six arms, ScreenSpot-v2 (554 icon crops, 408 whole screenshots), bench/reports/VISUAL.md: icon pool picture 0.208 words 0.175 chance 0.018 screenshot pool words 0.384 picture 0.215 screenshot to icon 0.108 — CLIP is a whole-image model, not a grounding model: it cannot point at one small icon inside a screen. within one screen MRR 0.908 (recall@10 is 1.000 there only because chance is 1.000 too, so MRR is the only honest number in that row). ScreenSpot-v2 instructions are all English, so this run does not test the bilingual claim and the report says so. Gates: 199 tests, ruff clean on 17 files, mypy clean on 16 (fresh cache), the repo's own pyright 0 errors on the 12 files touched.
1.8 KiB
1.8 KiB
VISUAL: can a library of wordless pictures be searched? (commit 7e1fe649f)
- generated 2026-10-08T00:37:50.278445+00:00
- corpus: OS-Copilot/ScreenSpot-v2 (icon instructions), 554 icon crops and 408 whole screenshots stored as separate memories, 427 crops with no OCR text
- encoder: Xenova/chinese-clip-vit-base-patch16 sha256 0fad978faa89b85f
- machine: 32 cpus, load [46.81, 44.0, 33.68]
- instructions containing Chinese: 0 (ScreenSpot-v2 is English, so this run does not test the bilingual claim)
| ask | pool | recall@1 | recall@5 | recall@10 | chance@10 | MRR | hit@10 lower 95% | p50 s | p95 s |
|---|---|---|---|---|---|---|---|---|---|
| words → icon crops | crops | 0.090 | 0.152 | 0.175 | 0.0181 | 0.118 | 0.146 | 0.1716 | 0.2096 |
| a sentence → icon crops (picture space) | crops | 0.081 | 0.155 | 0.208 | 0.0181 | 0.116 | 0.176 | 0.1399 | 0.1998 |
| the whole screenshot → icon crops | crops | 0.038 | 0.083 | 0.108 | 0.0181 | 0.058 | 0.085 | 0.1599 | 0.22 |
| words → whole screenshots | shots | 0.121 | 0.296 | 0.384 | 0.0245 | 0.196 | 0.345 | 0.1362 | 0.2038 |
| a sentence → whole screenshots (picture space) | shots | 0.049 | 0.146 | 0.215 | 0.0245 | 0.091 | 0.183 | 0.136 | 0.204 |
| a sentence → the icons of that one screen | own_screen | 0.827 | 1.000 | 1.000 | 1.0000 | 0.908 | 0.993 | 0.1358 | 0.1959 |
The words rows are what the product could do before any picture vector: if OCR found no words, there is nothing to match. The picture rows ask the same sentence, or the whole screenshot, of the CLIP space. The pools change the question: among every icon in the library, among whole screenshots (the granularity the store keeps), and among one screen's icons, which is the shape ScreenSpot itself poses. A number is only comparable inside its own pool, so chance is printed per row.