DocPTBench: Benchmarking End-to-End Photographed Document Parsing and Translation

If you find this project useful, please give us a star🌟.

DocPTBench is a benchmark designed specifically for real-world photographed documents, targeting both document parsing and document translation in challenging, realistic environments.

Unlike previous benchmarks built on clean-born digital documents, DocPTBench exposes models to:

• perspective distortion
• lighting variations / shadows
• motion blur
• physical folds & wrinkles
• noise and camera artifacts

This benchmark enables rigorous evaluation of both Document Parsing models and Multimodal LLMs (MLLMs) under practical conditions.

📈 Highlights from the Paper

(a): the results of MLLMs on English (En)-started parsing (P) and translation (T) tasks; (b): the counterpart on Chinese (Zh)-started tasks; (c): the results from document parsing expert models. Ori- refers to the original digital-born document and Photographed-is its photographed version. Text- indicates that only the textual content of the document image is used as the source-language input. Alower Edit distance indicates higher parsing quality, and a higher BLEU score reflects better translation fidelity.

• 📉 MLLMs an average parsing drops by 18% on photographed docs
• 📉 Expert models drop 25%
• 📉 Translation BLEU drops by 12%
• 🔧 Unwarping helps, but does not fully restore original quality
• 💡 CoT prompting greatly reduces instruction-following failures

---

🌟 Key Features

📷 1,381 Realistic Photographed Documents

Including both simulated and real-camera captures.

🌐 8 Language Pairs for Translation

En ↔ Zh / De / Fr / Ru and Zh ↔ En / De / Fr / Ru, all human-verified.

🖼 Three Document Conditions

Digital-Born (Original) → Photographed → Unwarping

🎯 End-to-End Evaluation

Supports both:

• Parsing-only models
• Unified end-to-end MLLMs

---

🏆 Document Parsing LeaderBoard

Comprehensive benchmarking of document parsing performance on DocPTBench. This table presents the full results for all specialized Expert Models and General MLLMs evaluated in our study. It contrasts model accuracy on three document types: pristine Original documents, Photographed images with real-world distortions, and geometrically corrected Unwarping images. The results quantify the performance degradation caused by photographic artifacts (subscript values next to Photographed scores) and the partial recovery achieved through unwarping (subscript values next to Unwarping scores). Lower Edit scores and higher TEDS scores indicate better performance.

Type	Model	Scene	OverallEdit↓		TextEdit↓		FormulaEdit↓		TableTEDS↑		TableEdit↓		Reading OrderEdit↓
			En	Zh	En	Zh	En	Zh	En	Zh	En	Zh	En	Zh
Expert Models
	HunyuanOCR	Original	9.6	10.9	3.1	6.6	26.4	28.8	94.0	95.9	4.7	3.3	4.3	4.9
		Photographed	22.4↓12.8	30.4↓19.5	17.0↓13.9	31.6↓25.0	34.3↓7.9	44.3↓15.5	74.0↓20.0	75.2↓20.7	22.9↓18.2	20.6↓17.3	15.4↓11.1	25.3↓20.4
		Unwarping	15.8↑6.6	22.8↑7.6	6.6↑10.4	19.3↑12.3	31.4↑2.9	44.0↑0.3	80.0↑6.1	84.9↑9.7	17.0↑5.9	12.5↑8.1	8.1↑7.3	15.4↑9.9
	PaddleOCR-VL	Original	10.5	12.6	4.1	6.2	24.1	31.6	88.0	92.1	9.3	6.2	4.5	6.3
		Photographed	37.5↓27.0	39.6↓27.0	29.4↓25.3	37.7↓31.5	46.5↓22.4	52.6↓21.0	54.2↓33.8	65.3↓26.8	44.4↓35.1	31.4↓25.2	28.8↓24.3	37.9↓31.6
		Unwarping	15.7↑21.8	22.0↑17.6	9.4↑20.0	17.6↑20.1	30.8↑15.7	41.5↑11.1	82.9↑28.7	83.2↑17.9	13.9↑30.5	13.5↑17.9	8.7↑20.1	15.4↑22.5
	MinerU2.5	Original	11.1	17.4	5.0	7.4	25.8	47.3	88.3	89.2	8.9	8.3	4.5	6.8
		Photographed	37.3↓26.2	47.4↓30.0	37.0↓32.0	53.6↓46.2	44.3↓18.5	62.0↓14.7	54.9↓33.4	59.8↓29.4	38.9↓30.0	33.5↓25.2	29.0↓24.5	40.3↓33.5
		Unwarping	17.3↑20.0	25.2↑22.2	13.1↑23.9	19.1↑34.5	31.9↑12.4	52.2↑9.8	79.2↑24.3	81.1↑21.3	15.7↑23.2	14.6↑18.9	8.3↑20.7	15.0↑25.3
	dots.ocr	Original	12.5	16.0	3.2	6.6	32.9	41.6	88.6	89.0	9.9	9.2	4.0	6.7
		Photographed	33.7↓21.2	37.3↓21.3	29.8↓26.6	35.8↓29.2	39.2↓6.3	54.4↓12.8	63.7↓24.9	67.6↓21.4	33.0↓23.1	27.1↓17.9	32.8↓28.8	31.8↓25.1
		Unwarping	16.3↑17.4	24.1↑13.2	8.3↑21.5	20.9↑14.9	32.2↑7.0	42.0↑12.4	80.2↑16.5	82.3↑14.7	16.9↑16.1	14.6↑12.5	7.9↑24.9	18.9↑12.9
	MonkeyOCR	Original	14.6	22.1	6.8	11.8	27.2	45.2	81.3	85.5	14.9	13.4	9.3	17.9
		Photographed	46.4↓31.8	52.8↓30.7	34.5↓27.7	43.9↓32.1	48.7↓21.5	61.6↓16.4	33.1↓48.2	37.4↓48.1	64.5↓49.6	61.5↓48.1	37.9↓28.6	44.1↓26.2
		Unwarping	18.8↑27.6	31.9↑20.9	12.5↑22.0	23.6↑20.3	32.1↑16.6	55.8↑5.8	77.2↑44.1	77.1↑39.7	17.2↑47.3	19.5↑42.0	13.5↑24.4	28.7↑15.4
	Dolphin	Original	20.5	31.3	9.2	20.4	44.7	60.6	76.1	66.9	19.3	28.2	8.8	11.6
		Photographed	57.5↓37.0	71.5↓40.2	54.9↓45.7	71.5↓51.1	65.6↓20.9	82.8↓22.2	33.0↓43.1	19.3↓47.6	67.9↓48.6	73.9↓45.7	46.2↓37.4	57.7↓46.1
		Unwarping	27.3↑30.2	45.5↑26.0	17.9↑37.0	36.9↑34.6	48.3↑17.3	75.1↑7.7	63.8↑30.8	48.6↑29.3	29.2↑38.7	42.5↑31.4	13.9↑32.3	27.3↑30.4
	olmOCR	Original	32.6	46.9	9.7	29.3	45.5	65.5	68.1	61.3	60.8	65.2	14.5	27.7
		Photographed	39.1↓6.5	46.1↑0.8	19.3↓9.6	27.2↑2.1	50.7↓5.2	66.9↓1.4	56.5↓11.6	56.9↓4.4	65.6↓4.8	66.0↓0.8	20.7↓6.2	24.4↑3.3
		Unwarping	31.4↑7.7	43.1↑3.0	9.6↑9.7	23.7↑3.5	40.0↑10.7	61.3↑5.6	65.8↑9.3	63.7↑6.8	62.7↑2.9	63.3↑2.7	13.4↑7.3	23.9↑0.5
	OCRFlux	Original	23.8	34.9	11.2	25.6	44.7	71.6	69.0	80.0	26.9	16.2	12.6	26.3
		Photographed	36.2↓12.4	45.8↓10.9	30.4↓19.2	40.4↓14.8	48.4↓3.7	81.1↓9.5	49.5↓19.5	54.3↓25.7	29.7↓2.8	29.7↓13.5	22.5↓9.9	32.1↓5.8
		Unwarping	23.6↑12.6	37.9↑7.9	11.8↑18.6	29.7↑10.7	42.5↑5.9	73.7↑7.4	68.1↑18.6	72.7↑18.4	27.6↑2.1	20.8↑8.9	12.7↑9.8	27.3↑4.8
	SmolDocling	Original	49.3	81.6	26.2	82.8	75.3	99.7	16.5	7.3	90.8	92.7	22.7	52.2
		Photographed	90.1↓40.8	93.7↓12.1	89.8↓63.6	99.2↓16.4	99.6↓24.3	99.9↓0.2	4.4↓12.1	2.4↓4.9	98.4↓7.6	98.8↓6.1	72.7↓50.0	75.9↓23.7
		Unwarping	65.2↑24.9	92.8↑0.9	45.6↑44.2	97.9↑1.3	92.8↑6.8	99.7↑0.2	25.9↑21.5	1.7↓0.7	90.0↑8.4	100.0↓1.2	38.6↑34.1	74.6↑1.3
	Nanonets-OCR	Original	28.3	29.5	13.4	23.1	51.8	54.6	76.8	79.4	34.3	20.1	13.5	20.0
		Photographed	38.6↓10.3	52.1↓22.6	21.0↓7.6	42.0↓18.9	48.1↑3.7	67.0↓12.4	58.5↓18.3	50.6↓28.8	64.1↓29.8	66.7↓46.6	21.4↓7.9	32.7↓12.7
		Unwarping	32.0↑6.6	44.4↑7.7	13.2↑7.8	30.2↑11.8	42.6↑5.5	65.6↑1.4	59.9↑1.4	59.8↑9.2	56.1↑8.0	56.1↑10.6	14.4↑7.0	25.6↑7.1
	DeepSeek-OCR	Original	13.4	18.1	4.6	9.7	28.5	43.3	82.6	89.0	13.8	8.8	6.7	10.5
		Photographed	54.4↓41.0	57.8↓39.7	56.7↓52.1	57.6↓47.9	54.4↓25.9	74.1↓30.8	28.0↓54.6	35.4↓53.6	64.7↓50.9	59.2↓50.4	41.7↓35.0	40.4↓29.9
		Unwarping	22.1↑32.3	33.5↑24.3	14.9↑41.8	29.4↑28.2	32.1↑22.3	58.8↑15.3	67.0↑39.0	75.8↑40.4	26.7↑38.0	20.9↑38.3	14.8↑26.9	24.9↑15.5
	olmOCR2	Original	16.1	26.7	4.8	18.5	39.2	54.3	83.7	78.5	12.3	16.5	8.1	17.4
		Photographed	27.8↓11.7	44.6↓17.9	22.0↓17.2	39.9↓21.4	44.6↓5.4	74.1↓19.8	67.6↓16.1	65.4↓13.1	24.6↓12.3	28.5↓12.0	19.9↓11.8	36.0↓18.6
		Unwarping	17.5↑10.3	37.2↑7.4	7.3↑14.7	32.9↑7.0	37.5↑7.1	66.7↑7.4	81.9↑14.3	77.2↑11.8	14.3↑10.3	19.1↑9.4	11.0↑8.9	30.2↑5.8
	Nanonets-OCR2	Original	26.6	34.9	19.4	34.3	60.0	68.0	81.5	82.5	15.5	17.9	11.6	19.4
		Photographed	34.2↓7.6	46.1↓11.2	25.5↓6.1	44.6↓10.3	69.0↓9.0	76.4↓8.4	70.7↓10.8	66.0↓16.5	22.8↓7.3	31.9↓14.0	19.5↓7.9	31.4↓12.0
		Unwarping	30.6↑3.6	40.0↑6.1	21.1↑4.4	32.6↑12.0	65.3↑3.7	77.3↓0.9	71.9↑1.2	73.1↑7.1	24.8↓2.0	18.5↑13.4	17.5↑2.0	25.2↑6.2
General MLLMs	Qwen2.5-VL-72B	Original	21.4	26.1	9.2	18.0	31.5	43.4	82.9	83.9	34.1	26.2	10.6	16.8
		Photographed	41.5↓20.1	57.0↓30.9	36.2↓27.0	56.6↓38.6	42.2↓10.7	61.8↓18.4	57.0↓25.9	55.5↓28.4	59.6↓25.5	58.2↓32.0	28.1↓17.5	51.3↓34.5
		Unwarping	24.0↑17.5	41.4↑15.6	11.1↑25.1	42.7↑13.9	29.9↑12.3	48.4↑13.4	77.4↑20.4	76.1↑20.6	42.7↑16.9	34.9↑23.3	12.3↑15.8	39.7↑11.6
	Gemini2.5-Pro	Original	14.8	21.2	5.5	16.8	35.6	43.9	85.8	86.4	13.0	11.9	4.9	12.1
		Photographed	18.2↓3.4	30.4↓9.2	9.8↓4.3	27.7↓10.9	37.1↓1.5	56.8↓12.9	81.3↓4.5	82.9↓3.5	14.6↓1.6	13.7↓1.8	11.2↓6.3	23.6↓11.5
		Unwarping	16.9↑1.3	27.3↑3.1	9.2↑0.6	20.8↑6.9	35.3↑1.8	57.0↓0.2	83.4↑2.1	85.9↑3.0	13.1↑1.5	11.8↑1.9	10.0↑1.2	19.8↑3.8
	Doubao-1.6-v	Original	22.5	29.3	16.2	27.6	31.2	47.2	66.6	76.3	31.9	24.5	10.8	17.9
		Photographed	54.7↓32.2	55.4↓26.1	60.6↓44.4	58.2↓30.6	51.5↓20.3	61.1↓13.9	27.6↓39.0	37.9↓38.4	67.0↓35.1	61.9↓37.4	39.7↓28.9	40.2↓22.3
		Unwarping	30.0↑24.7	42.5↑12.9	23.8↑36.8	41.8↑16.4	34.5↑17.0	56.4↑4.7	55.7↑28.1	60.8↑22.9	44.9↑22.1	42.4↑19.5	16.7↑23.0	29.5↑10.7
	Qwen-VL-Max	Original	16.6	26.5	5.2	20.5	32.9	44.0	84.2	86.7	22.0	23.7	6.5	17.7
		Photographed	27.7↓11.1	42.7↓16.2	15.9↓10.7	41.5↓21.0	41.8↓8.9	57.2↓13.2	71.1↓13.1	71.6↓15.1	36.3↓14.3	38.0↓14.3	16.8↓10.3	34.4↓16.7
		Unwarping	19.0↑8.7	32.6↑10.1	6.8↑9.1	32.1↑9.4	33.8↑8.0	48.5↑8.7	81.3↑10.2	83.3↑11.7	26.5↑9.8	22.0↑16.0	9.0↑7.8	27.8↑6.6
	GLM-4.5v	Original	25.5	32.0	16.1	27.7	43.8	51.8	74.0	77.4	26.9	30.5	15.4	17.9
		Photographed	36.7↓11.2	49.6↓17.6	26.2↓10.1	47.7↓20.0	49.9↓6.1	66.2↓14.4	58.9↓15.1	54.0↓23.4	43.5↓16.6	49.0↓18.5	27.3↓11.9	35.7↓17.8
		Unwarping	23.9↑12.8	36.9↑12.7	13.1↑13.1	37.7↑10.0	39.0↑10.9	53.5↑12.7	73.8↑14.9	75.6↑21.6	26.9↑16.6	28.7↑20.3	16.5↑10.8	27.7↑8.0
	Kimi-VL	Original	36.5	38.7	17.2	22.0	48.6	52.2	57.1	67.8	65.9	62.5	14.3	18.1
		Photographed	69.6↓33.1	68.7↓30.0	66.0↓48.8	63.5↓41.5	75.5↓26.9	82.6↓30.4	16.4↓40.7	22.9↓44.9	85.4↓19.5	82.2↓19.7	51.6↓37.3	46.7↓28.6
		Unwarping	41.1↑28.5	50.7↑18.0	26.3↑39.7	38.5↑25.0	50.4↑25.1	68.8↑13.8	55.4↑39.0	62.3↑39.4	65.4↑20.0	65.0↑17.2	22.1↑29.5	30.7↑16.0

---

🏆 Document Translation LeaderBoard

End-to-End translation performance on digital-born (Original) documents. We quantify the performance gap between a model's pure text translation capability (Text-only baseline) and its end-to-end performance on document images. It further compares a Simple prompting strategy against a CoT approach to evaluate how decoupling perception and translation affects accuracy. Results for the other six language pairs are detailed in the supplementary material.

Type	Model	Input	En-Zh				Zh-En
			BLEU	chrF	METEOR	STEDS	BLEU	chrF	METEOR	STEDS
Open Source	Qwen3-VL-4B	Text	49.61	56.87	66.74	94.35	50.20	72.82	64.91	94.24
		Original-Simple	32.11↓17.50	40.22↓16.65	47.49↓19.25	64.55↓29.80	28.31↓21.89	48.72↓24.10	40.44↓24.47	68.41↓25.83
		Original-CoT	36.86↑4.75	45.17↑4.95	53.97↑6.48	68.83↑4.28	34.84↑6.53	57.29↑8.57	48.75↑8.31	66.14↓2.27
	Qwen2.5-VL-3B	Text	48.60	55.39	63.91	81.59	45.29	66.13	57.55	87.35
		Original-Simple	18.18↓30.42	25.65↓29.74	27.42↓36.49	59.02↓22.57	15.20↓30.09	23.73↓42.40	20.78↓36.77	60.87↓26.48
		Original-CoT	19.37↑1.19	28.85↑3.20	32.09↑4.67	49.57↓9.45	18.50↑3.30	35.56↑11.83	28.98↑8.20	48.24↓12.63
	InternVL3-2B	Text	48.25	54.29	62.48	89.42	33.54	50.01	43.78	84.94
		Original-Simple	10.87↓37.38	17.33↓36.96	18.91↓43.57	55.90↓33.52	7.27↓26.27	11.63↓38.38	10.38↓33.40	57.83↓27.11
		Original-CoT	19.21↑8.34	28.07↑10.74	32.91↑14.00	55.16↓0.74	22.07↑14.80	46.01↑34.38	36.06↑25.68	59.16↑1.33
	InternVL3.5-2B	Text	57.49	63.14	72.23	94.29	48.46	69.48	61.02	92.18
		Original-Simple	25.43↓32.06	34.62↓28.52	40.15↓32.08	64.44↓29.85	8.42↓40.04	11.04↓58.44	10.52↓50.50	65.03↓27.15
		Original-CoT	31.42↑5.99	41.25↑6.63	48.69↑8.54	65.14↑0.70	28.28↑19.86	50.16↑39.12	41.75↑31.23	61.86↓3.17
	HunyuanOCR	Original-self	16.77	24.91	29.17	58.88	12.49	26.25	19.04	56.09
		Photographed-self	13.49↓3.28	21.86↓3.05	25.68↓3.49	53.22↓5.66	10.50↓1.99	25.02↓1.23	18.63↓0.41	50.72↓5.27
Closed Source	Gemini2.5-Pro	Text	60.07	66.54	76.39	92.90	53.62	76.01	70.06	91.23
		Original-Simple	44.34↓15.73	53.83↓12.71	64.97↓11.42	71.77↓21.13	37.96↓15.66	67.45↓8.56	58.04↓12.02	65.75↓25.48
		Original-CoT	44.41↑0.07	53.94↑0.11	65.68↑0.71	75.05↑3.28	42.81↑4.85	69.62↑2.17	61.67↑3.63	75.37↑9.62
		Photographed-Simple	43.72↓0.62	53.77↓0.06	63.68↓1.29	71.82↑0.05	32.88↓5.08	62.95↓4.50	52.24↓5.80	63.42↓2.33
		Photographed-CoT	43.88↓0.53	53.88↓0.06	64.06↓1.62	75.18↑0.13	34.89↓7.92	61.59↓8.03	51.88↓9.79	70.26↓5.11
	Qwen-VL-Max	Text	69.41	74.05	82.81	96.91	54.33	75.19	67.35	92.19
		Original-Simple	41.04↓28.37	50.81↓23.24	59.77↓23.04	72.76↓24.15	36.29↓18.04	61.03↓14.16	50.40↓16.95	71.68↓20.51
		Original-CoT	47.60↑6.56	55.70↑4.89	64.10↑4.33	72.67↓0.09	42.28↑5.99	66.05↑5.02	56.44↑6.04	69.68↓2.00
		Photographed-Simple	27.53↓13.51	37.25↓13.56	43.81↓15.96	69.02↓3.74	21.81↓14.48	45.93↓15.10	34.44↓15.96	64.96↓6.72
		Photographed-CoT	37.44↓10.16	46.76↓8.94	54.99↓9.11	68.24↓4.43	30.64↓11.64	54.88↓11.17	44.43↓12.01	64.16↓5.52
	GLM-4.5v	Text	62.53	68.38	77.84	95.57	55.51	75.62	68.56	92.84
		Original-Simple	42.14↓20.39	51.20↓17.18	60.82↓17.02	73.72↓21.85	39.02↓16.49	62.67↓12.95	53.10↓15.46	74.34↓18.50
		Original-CoT	45.90↑3.76	55.09↑3.89	64.91↑4.09	73.14↓0.58	42.34↑3.32	66.92↑4.25	57.48↑4.38	72.43↓1.91
		Photographed-Simple	31.03↓11.11	41.02↓10.18	47.41↓13.41	71.21↓2.51	24.82↓14.20	46.42↓16.25	37.45↓15.65	60.44↓13.90
		Photographed-CoT	37.48↓8.42	46.72↓8.37	54.39↓10.52	70.94↓2.20	29.88↓12.46	53.71↓13.21	44.15↓13.33	62.60↓9.83
	Kimi-VL	Text	67.95	72.45	81.78	97.34	60.76	78.64	73.47	95.61
		Original-Simple	38.20↓29.75	47.17↓25.28	55.14↓26.64	70.38↓26.96	32.07↓28.69	54.72↓23.92	44.93↓28.54	69.85↓25.76
		Original-CoT	42.36↑4.16	50.94↑3.77	58.68↑3.54	68.66↓1.72	42.63↑10.56	64.24↑9.52	55.75↑10.82	69.03↓0.82
		Photographed-Simple	9.16↓29.04	15.97↓31.20	20.51↓34.63	49.05↓21.33	9.15↓22.92	27.77↓26.95	18.52↓26.41	50.99↓18.86
		Photographed-CoT	12.07↓30.29	19.17↓31.77	23.46↓35.22	52.42↓16.24	15.78↓26.85	34.88↓29.36	26.49↓29.26	49.07↓19.96
	Doubao-1.6-v	Text	54.92	62.59	72.26	87.26	46.15	71.22	62.51	83.70
		Original-Simple	39.29↓15.63	49.73↓12.86	59.29↓12.97	69.80↓17.46	34.31↓11.84	61.94↓9.28	51.50↓11.01	70.99↓12.71
		Original-CoT	41.61↑2.32	51.09↑1.36	61.32↑2.03	71.52↑1.72	36.98↑2.67	64.47↑2.53	54.26↑2.76	71.98↑0.99
		Photographed-Simple	35.36↓3.93	46.47↓3.26	53.60↓5.69	66.46↓3.34	26.88↓7.43	53.62↓8.32	42.58↓8.92	63.27↓7.72
		Photographed-CoT	39.61↓2.00	49.61↓1.48	57.88↓3.44	66.70↓4.82	29.91↓7.07	56.52↓7.95	45.97↓8.29	63.53↓8.45

---

🖼️ Example Input & Output

Refer to the appendix of the paper.

---

🧪 Evaluation

Document Parsing

Refer to the parsing.md for evaluation details.

Document Translation

Refer to the translation.md for evaluation details.

---

🧩 Supported Model Families

📘 Document Parsing Models

• PaddleOCR-VL
• MinerU2.5
• dots.ocr
• MonkeyOCR
• DeepSeek-OCR
• olmOCR and olmOCR2
• Dolphin
• OCRFlux
• SmolDocling
• Nanonets-OCR and Nanonets-OCR2
• HunyuanOCR

🤖 MLLMs (Closed-Source)

• Gemini2.5 Pro
• Qwen-VL-Max
• Kimi-VL
• GLM-4.5v
• Doubao 1.6-v
• Gemini3 Pro

🔓 Open-Source Lightweight Models

• Qwen3-VL-4B
• Qwen2.5-VL-3B
• InternVL3-2B
• InternVL3.5-2B
• Qwen3-VL-235B

---

📚 Citation

If you use DocPTBench, please cite:

@misc{docptbench2025,
  title={DocPTBench: Benchmarking End-to-End Photographed Document Parsing and Translation},
  author={Yongkun Du and Pinxuan Chen and Xuye Ying and Zhineng Chen},
  year={2025},
  eprint={2511.18434},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2511.18434}
}

Additionally, we encourage you to cite the following papers:

@misc{ouyang2024omnidocbenchbenchmarkingdiversepdf,
      title={OmniDocBench: Benchmarking Diverse PDF Document Parsing with Comprehensive Annotations}, 
      author={Linke Ouyang and Yuan Qu and Hongbin Zhou and Jiawei Zhu and Rui Zhang and Qunshu Lin and Bin Wang and Zhiyuan Zhao and Man Jiang and Xiaomeng Zhao and Jin Shi and Fan Wu and Pei Chu and Minghao Liu and Zhenxiang Li and Chao Xu and Bo Zhang and Botian Shi and Zhongying Tu and Conghui He},
      year={2024},
      eprint={2412.07626},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2412.07626}, 
}

---

🙏 Acknowledgments

DocPTBench is developed based on OmniDocBench. Thanks for their awesome work!