Add paddle-calfa-small results

README.md

@@ -267,9 +267,10 @@ Document structure-aware models output one complete text line per box, scored di
 | [abbyy](https://www.abbyy.com/ocr-sdk/) | 43.09% | 42.92% | 42.78% | 42.74% |
 | [google-vision](https://docs.cloud.google.com/vision/docs/ocr) | 43.46% | 43.49% | 43.99% | 43.99% |
 | [portmind](https://github.com/portmind/armenian-ocr) | 24.74% | 24.67% | 24.63% | 24.60% |
-| [**ocarm-v1***](https://huggingface.co/datasets/RLALT/ACoPPer/blob/main/README.md#ocarm) | **4.15%** | **4.07%** | **3.92%** | **3.90%** |
+| [ocarm-v1*](https://huggingface.co/datasets/RLALT/ACoPPer/blob/main/README.md#ocarm) | 4.15% | 4.07% | 3.92% | 3.90% |
 | [hy-open-ocr](https://github.com/calfa-co/hye-open-ocr) | 11.49% | 11.28% | 10.82% | 10.76% |
-| [paddle-calfa-tiny](https://github.com/calfa-co/hye-open-ocr) | 4.87% | 4.72% | 4.58% | 4.54% |
+| [paddle-calfa-tiny](https://github.com/calfa-co/hye-paddle) | 4.87% | 4.72% | 4.58% | 4.54% |
+| [**paddle-calfa-small**](https://github.com/calfa-co/hye-paddle) | **3.34%** | **3.20%** | **3.07%** | **3.04%** |
 
 ### Document structure-independent models
 Document structure-independent models output individual word boxes, which are spatially grouped into lines before scoring. These models return one bounding box + recognized text per individual word. Before scoring, script spatially groups words into lines using vertical-overlap and horizontal-gap heuristics: words that share a line are concatenated in reading order into a single predicted row, which is then matched against whichever ground-truth box(es) it falls inside and compared via character edit distance. Because line breaks are reconstructed rather than given, document structure-independent scores also depend on how well that grouping heuristic reassembles lines — a good recognizer can still lose points here if word boxes are noisy enough to throw off the grouping.
@@ -282,9 +283,10 @@ Document structure-independent models output individual word boxes, which are sp
 | [abbyy](https://www.abbyy.com/ocr-sdk/) | 37.32% | 37.15% | 36.91% | 36.86% |
 | [google-vision](https://docs.cloud.google.com/vision/docs/ocr) | 7.52% | 7.46% | 7.33% | 7.30% |
 | [portmind](https://github.com/portmind/armenian-ocr) | 6.76% | 6.71% | 6.58% | 6.56% |
-| [**ocarm-v1***](https://huggingface.co/datasets/RLALT/ACoPPer/blob/main/README.md#ocarm) | **4.14%** | **4.07%** | **3.94%** | **3.92%** |
+| [ocarm-v1*](https://huggingface.co/datasets/RLALT/ACoPPer/blob/main/README.md#ocarm) | 4.14% | 4.07% | 3.94% | 3.92% |
 | [hy-open-ocr](https://github.com/calfa-co/hye-open-ocr) | 11.49% | 11.28% | 10.82% | 10.76% |
-| [paddle-calfa-tiny](https://github.com/calfa-co/hye-open-ocr) | 4.87% | 4.72% | 4.58% | 4.54% |
+| [paddle-calfa-tiny](https://github.com/calfa-co/hye-paddle) | 4.87% | 4.72% | 4.58% | 4.54% |
+| [**paddle-calfa-small**](https://github.com/calfa-co/hye-paddle) | **3.34%** | **3.20%** | **3.07%** | **3.04%** |
 
 **Additional models — [DeepSeek-OCR](https://huggingface.co/deepseek-ai/DeepSeek-OCR), [Qwen-VL](https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct), and [Chandra](https://github.com/datalab-to/chandra) — were also evaluated on this dataset but produced substantially higher error rates (50%+ CER) and are not included in the tables above.**