One example where the model fails on tables

Take the big table page 11 of your paper.

Process it, it will not detect the last two columns.

<|ref|>table<|/ref|><|det|>[[120, 210, 874, 725]]<|/det|>
<|ref|>table_caption<|/ref|><|det|>[[113, 99, 880, 201]]<|/det|>
Table 3 | We use OmniDocBench [27] to test the performance of DeepSeek-OCR on real document parsing tasks. All metrics in the table are edit distances, where smaller values indicate better performance. "Tokens" represents the average number of vision tokens used per page, and \(\mathrm{^{i + 200dpi}}\) means using fitz to interpolate the original image to 200dpi. For the DeepSeek-OCR model, the values in parentheses in the "Tokens" column represent valid vision tokens, calculated according to Equation 1.

ModelTokensEnglishChineseoveralltextformulatableorderoveralltextformulaPipline ModelsDolphin [11]-0.3560.3520.4650.2580.350.440.440.604Marker [1]-0.2960.0850.3740.6090.1160.4970.2930.688Mathpix [2]-0.1910.1050.3060.2430.1080.3640.3810.454MinerU-2.1.1 [34]-0.1620.0720.3130.1660.0970.2440.1110.581MonkeyOCR-1.2B [18]-0.1540.0620.2950.1640.0940.2630.1790.464PPstructure-v3 [9]-0.1520.0730.2950.1620.0770.2230.1360.535End-to-end ModelsNougat [6]23520.4520.3650.4880.5720.3820.9730.9980.941SmolDocking [25]3920.4930.2620.7530.7290.2270.8160.8380.997InternVL2-76B [8]67900.440.3530.5430.5470.3170.4430.290.701Qwen2.5-VL-7B [5]39490.3160.1510.3760.5980.1380.3990.2430.5OLMOCR [28]39490.3260.0970.4550.6080.1450.4690.2930.655GOT-OCR2.0 [38]2560.2870.1890.3600.4590.1410.4110.3150.528OCRFlux-3B [3]39490.2380.1120.4470.2690.1260.3490.2560.716GPT4o [26]-0.2330.1440.4250.2340.1280.3990.4090.606InternVL3-78B [42]67900.2180.1170.380.2790.0950.2960.210.533Qwen2.5-VL-72B [5]39490.2140.0920.3150.3410.1060.2610.180.434dots.ocr [30]39490.1820.1370.3200.1660.1820.2610.2290.468Gemini2.5-Pro [4]-0.1480.0550.3560.130.0490.2120.1680.439MinerU2.0 [34]67900.1330.0450.2730.150.0660.2380.1150.506dots.ocr+200dpi [30]55450.1250.0320.3290.0990.040.160.0660.416DeepSeek-OCR (end2end)Tiny640.3860.3730.4690.4220.2830.3610.3070.635Small1000.2210.1420.3730.2420.1250.2840.240.53Base256(182)0.1370.0540.2670.1630.0640.240.2050.474Large400(285)0.1380.0540.2770.1520.0670.2080.1430.461Gundam7950.1270.0430.2690.1340.0620.1810.0970.432Gundam-M+200dpi18530.1230.0490.2420.1470.0560.1570.0870.377