←back to thread

DeepSeek OCR

(github.com)
990 points pierre | 4 comments | 20 Oct 25 06:26 UTC | HN request time: 0.201s | source
1. bugglebeetle ◴[20 Oct 25 07:01 UTC] No.45640730[source]▶
Looks great, but looking at the benchmark, can’t help but think about how crazy good dots-ocr is as a model. Too bad they’re not as open as the Deepseek team because its so crazy good and would love to know how it was trained.
replies(2): >>45640765 #>>45640789 #
2. rfoo ◴[20 Oct 25 07:09 UTC] No.45640765[source]▶
If you look you'd notice that it's the same Haoran Wei behind DeepSeek-OCR and GOT-OCR2.0 :p
replies(1): >>45646398 #
3. bethekind ◴[20 Oct 25 07:13 UTC] No.45640789[source]▶
Did we read the same graph? DeepSeek Gundam 200 dpi appeared to get similar perf as dots-ocr, but with less tokens needed. The x axis is inverted, descending with distance from the origin.
4. bugglebeetle ◴[20 Oct 25 17:13 UTC] No.45646398[source]▶
Oh you’re right! Good catch!