AIStoryBoard/python/PaddleOCR-main/docs/version3.x/algorithm/PP-OCRv5/PP-OCRv5.en.md

Introduction to PP-OCRv5

PP-OCRv5 is the new generation text recognition solution of PP-OCR, focusing on multi-scenario and multi-text type recognition. In terms of text types, PP-OCRv5 supports 5 major mainstream text types: Simplified Chinese, Chinese Pinyin, Traditional Chinese, English, and Japanese. For scenarios, PP-OCRv5 has upgraded recognition capabilities for challenging scenarios such as complex Chinese and English handwriting, vertical text, and uncommon characters. On internal complex evaluation sets across multiple scenarios, PP-OCRv5 achieved a 13 percentage point end-to-end improvement over PP-OCRv4.

Key Metrics

1. Text Detection Metrics

Model	Handwritten Chinese	Handwritten English	Printed Chinese	Printed English	Traditional Chinese	Ancient Text	Japanese	General Scenario	Pinyin	Rotation	Distortion	Artistic Text	Average
PP-OCRv5_server_det	0.803	0.841	0.945	0.917	0.815	0.676	0.772	0.797	0.671	0.8	0.876	0.673	0.827
PP-OCRv4_server_det	0.706	0.249	0.888	0.690	0.759	0.473	0.685	0.715	0.542	0.366	0.775	0.583	0.662
PP-OCRv5_mobile_det	0.744	0.777	0.905	0.910	0.823	0.581	0.727	0.721	0.575	0.647	0.827	0.525	0.770
PP-OCRv4_mobile_det	0.583	0.369	0.872	0.773	0.663	0.231	0.634	0.710	0.430	0.299	0.715	0.549	0.624

Compared to PP-OCRv4, PP-OCRv5 shows significant improvement in all detection scenarios, especially in handwriting, ancient texts, and Japanese detection capabilities.

2. Text Recognition Metrics

Evaluation Set Category	Handwritten Chinese	Handwritten English	Printed Chinese	Printed English	Traditional Chinese	Ancient Text	Japanese	Confusable Characters	General Scenario	Pinyin	Vertical Text	Artistic Text	Weighted Average
PP-OCRv5_server_rec	0.5807	0.5806	0.9013	0.8679	0.7472	0.6039	0.7372	0.5946	0.8384	0.7435	0.9314	0.6397	0.8401
PP-OCRv4_server_rec	0.3626	0.2661	0.8486	0.6677	0.4097	0.3080	0.4623	0.5028	0.8362	0.2694	0.5455	0.5892	0.5735
PP-OCRv5_mobile_rec	0.4166	0.4944	0.8605	0.8753	0.7199	0.5786	0.7577	0.5570	0.7703	0.7248	0.8089	0.5398	0.8015
PP-OCRv4_mobile_rec	0.2980	0.2550	0.8398	0.6598	0.3218	0.2593	0.4724	0.4599	0.8106	0.2593	0.5924	0.5555	0.5301

A single model can cover multiple languages and text types, with recognition accuracy significantly ahead of previous generation products and mainstream open-source solutions.

PP-OCRv5 Demo Examples

More Demos

Reference Data for Inference Performance

Test Environment:

• NVIDIA Tesla V100
• Intel Xeon Gold 6271C
• PaddlePaddle 3.0.0

Tested on 200 images (including both general and document images). During testing, images are read from disk, so the image reading time and other associated overhead are also included in the total time consumption. If the images are preloaded into memory, the average time per image can be further reduced by approximately 25 ms.

Unless otherwise specified:

• PP-OCRv4_mobile_det and PP-OCRv4_mobile_rec models are used.
• Document orientation classification, image correction, and text line orientation classification are not used.
• text_det_limit_type is set to "min" and text_det_limit_side_len to 736.

1. Comparison of Inference Performance Between PP-OCRv5 and PP-OCRv4

Config	Description
v5_mobile	Uses PP-OCRv5_mobile_det and PP-OCRv5_mobile_rec models.
v4_mobile	Uses PP-OCRv4_mobile_det and PP-OCRv4_mobile_rec models.
v5_server	Uses PP-OCRv5_server_det and PP-OCRv5_server_rec models.
v4_server	Uses PP-OCRv4_server_det and PP-OCRv4_server_rec models.

GPU

Configuration	Avg. Time per Image (s)	Avg. Characters Predicted per Second	Avg. CPU Utilization (%)	Peak RAM Usage (MB)	Avg. RAM Usage (MB)	Avg. GPU Utilization (%)	Peak VRAM Usage (MB)	Avg. VRAM Usage (MB)
v5_mobile	0.62	1054.23	106.35	1829.36	1521.92	17.42	4190.00	3114.02
v4_mobile	0.29	2062.53	112.21	1713.10	1456.14	26.53	1304.00	1166.68
v5_server	0.74	878.84	105.68	1899.80	1569.46	34.39	5402.00	4683.93
v4_server	0.47	1322.06	108.06	1773.10	1518.94	55.25	6760.67	5788.02

CPU

Configuration	Avg. Time per Image (s)	Avg. Characters Predicted per Second	Avg. CPU Utilization (%)	Peak RAM Usage (MB)	Avg. RAM Usage (MB)
v5_mobile	1.75	371.82	965.89	2219.98	1830.97
v4_mobile	1.37	444.27	1007.33	2090.53	1797.76
v5_server	4.34	149.98	990.24	4020.85	3137.20
v4_server	5.42	115.20	999.03	4018.35	3105.29

Note: PP-OCRv5 uses a larger dictionary in the recognition model, which increases inference time and causes slower performance compared to PP-OCRv4.

2. Impact of Auxiliary Features on PP-OCRv5 Inference Performance

Config	Description
base	No document orientation classification, no image correction, no text line orientation classification.
with_textline	Includes text line orientation classification only.
with_all	Includes document orientation classification, image correction, and text line orientation classification.

GPU

Configuration	Avg. Time per Image (s)	Avg. Characters Predicted per Second	Avg. CPU Utilization (%)	Peak RAM Usage (MB)	Avg. RAM Usage (MB)	Avg. GPU Utilization (%)	Peak VRAM Usage (MB)	Avg. VRAM Usage (MB)
base	0.62	1054.23	106.35	1829.36	1521.92	17.42	4190.00	3114.02
with_textline	0.64	1012.32	106.37	1867.69	1527.42	19.16	4198.00	3115.05
with_all	1.09	562.99	105.67	2381.53	1792.48	10.77	2480.00	2065.54

CPU

Configuration	Avg. Time per Image (s)	Avg. Characters Predicted per Second	Avg. CPU Utilization (%)	Peak RAM Usage (MB)	Avg. RAM Usage (MB)
base	1.75	371.82	965.89	2219.98	1830.97
with_textline	1.87	347.61	972.08	2232.38	1822.13
with_all	3.13	195.25	828.37	2751.47	2179.70

Note: Auxiliary features such as image unwarping can impact inference accuracy. More features do not necessarily yield better results and may increase resource usage.

3. Impact of Input Scaling Strategy in Text Detection Module on PP-OCRv5 Inference Performance

Config	Description
mobile_min_1280	Uses min limit type and text_det_limit_side_len=1280 with PP-OCRv5_mobile models.
mobile_min_736	Same as default, min, side_len=736.
mobile_max_960	Uses max limit type and side_len=960.
mobile_max_640	Uses max limit type and side_len=640.
server_min_1280	Uses min, side_len=1280 with PP-OCRv5_server models.
server_min_736	Same as default, min, side_len=736.
server_max_960	Uses max, side_len=960.
server_max_640	Uses max, side_len=640.

GPU

Configuration	Avg. Time per Image (s)	Avg. Characters Predicted per Second	Avg. CPU Utilization (%)	Peak RAM Usage (MB)	Avg. RAM Usage (MB)	Avg. GPU Utilization (%)	Peak VRAM Usage (MB)	Avg. VRAM Usage (MB)
mobile_min_1280	0.66	985.77	109.52	1878.74	1536.43	18.01	4050.00	3407.33
mobile_min_736	0.62	1054.23	106.35	1829.36	1521.92	17.42	4190.00	3114.02
mobile_max_960	0.52	1206.68	104.01	1795.27	1484.73	18.66	2490.00	2173.91
mobile_max_640	0.45	1353.49	103.32	1728.91	1470.64	18.55	2378.00	1998.62
server_min_1280	0.86	759.10	107.81	1876.31	1572.20	37.33	10368.00	8287.41
server_min_736	0.74	878.84	105.68	1899.80	1569.46	34.39	5402.00	4683.93
server_max_960	0.64	988.85	103.61	1831.31	1544.26	30.29	2929.33	2079.90
server_max_640	0.57	1036.90	102.89	1838.36	1532.50	28.91	3153.33	2743.40

4. Pipeline benchmark data

Click to expand/collapse the table

Pipeline configuration	Hardware	Avg. inference time (s)	Peak CPU utilization (%)	Avg. CPU utilization (%)	Peak host memory (MB)	Avg. host memory (MB)	Peak GPU utilization (%)	Avg. GPU utilization (%)	Peak device memory (MB)	Avg. device memory (MB)
OCR-default	Intel 6271C	3.97	1015.40	917.61	4381.22	3457.78	N/A	N/A	N/A	N/A
	Intel 8350C	3.79	1022.50	921.68	4675.46	3585.96	N/A	N/A	N/A	N/A
	Intel 8350C + A100	0.65	113.50	102.48	2240.15	1868.44	47	19.60	7612.00	6634.15
	Intel 6271C + V100	1.06	114.90	103.05	2142.66	1791.43	72	20.01	5516.00	4812.81
	Intel 8563C + H20	0.65	108.90	101.95	2456.05	2080.26	100	36.52	6736.00	6017.05
	Intel 8350C + A10	0.74	115.90	102.22	2352.88	1993.39	100	25.56	6762.00	6039.93
	Intel 6271C + T4	1.17	107.10	101.78	2361.88	1986.61	100	51.11	5282.00	4585.10
OCR-nopp-mobile	Intel 6271C	1.39	1019.60	1007.69	2178.12	1873.73	N/A	N/A	N/A	N/A
	Intel 8350C	1.15	1015.70	1006.87	2184.91	1916.85	N/A	N/A	N/A	N/A
	Hygon 7490 + P800	0.35	110.80	103.77	2022.49	1808.11	N/A	N/A	N/A	N/A
	Intel 8350C + A100	0.27	110.90	103.80	1762.36	1525.04	31	19.30	4328.00	3356.30
	Intel 6271C + V100	0.55	113.80	103.68	1728.02	1470.52	38	18.59	4198.00	3199.12
	Intel 8563C + H20	0.22	111.90	103.99	2073.88	1876.14	32	20.25	4386.00	3435.86
	Intel 8350C + A10	0.31	119.90	104.24	2037.38	1771.06	52	32.74	3446.00	2733.21
	M4	6.51	147.30	106.24	3550.58	3236.75	N/A	N/A	N/A	N/A
	Intel 6271C + T4	0.46	111.90	103.11	2035.38	1742.39	65	46.77	3968.00	2991.91
OCR-nopp-server	Intel 6271C	3.00	1016.00	1004.87	4445.46	3179.86	N/A	N/A	N/A	N/A
	Intel 8350C	3.23	1010.70	1002.63	4175.39	3137.58	N/A	N/A	N/A	N/A
	Intel 8350C + A100	0.34	110.90	103.30	1904.99	1591.10	57	32.29	7494.00	6551.47
	Intel 6271C + V100	0.69	108.90	102.95	1808.30	1568.64	72	35.30	5410.00	4741.18
	Intel 8563C + H20	0.38	109.40	102.34	2100.00	1863.73	100	50.18	6614.00	5926.51
	Intel 8350C + A10	0.41	109.00	103.18	2055.21	1845.14	100	47.15	6654.00	5951.22
	Intel 6271C + T4	0.82	104.40	101.73	1906.88	1689.69	100	76.41	5178.00	4502.64
OCR-nopp-min736-mobile	Intel 6271C	1.41	1020.10	1008.14	2184.16	1911.86	N/A	N/A	N/A	N/A
	Intel 8350C	1.20	1015.70	1007.08	2254.04	1935.18	N/A	N/A	N/A	N/A
	Hygon 7490 + P800	0.36	112.90	104.29	2174.58	1827.67	N/A	N/A	N/A	N/A
	Intel 8350C + A100	0.27	113.90	104.48	1717.55	1529.77	30	19.54	4328.00	3388.44
	Intel 6271C + V100	0.57	118.80	104.45	1693.10	1470.74	40	19.83	4198.00	3206.91
	Intel 8563C + H20	0.22	113.40	104.66	2037.13	1797.10	31	20.64	4384.00	3427.91
	Intel 8350C + A10	0.31	119.30	106.05	1879.15	1732.39	49	30.40	3446.00	2751.08
	M4	6.39	124.90	107.16	3578.98	3209.90	N/A	N/A	N/A	N/A
	Intel 6271C + T4	0.47	109.60	103.26	1961.40	1742.95	60	44.26	3968.00	3002.81
OCR-nopp-min736-server	Intel 6271C	3.26	1068.50	1004.96	4582.52	3135.68	N/A	N/A	N/A	N/A
	Intel 8350C	3.52	1010.70	1002.33	4723.23	3209.27	N/A	N/A	N/A	N/A
	Intel 8350C + A100	0.35	108.90	103.94	1703.65	1485.50	60	35.54	7492.00	6576.97
	Intel 6271C + V100	0.71	110.80	103.54	1800.06	1559.28	78	36.65	5410.00	4741.55
	Intel 8563C + H20	0.40	110.20	102.75	2012.64	1843.45	100	55.74	6614.00	5940.44
	Intel 8350C + A10	0.44	114.90	103.87	2002.72	1773.17	100	49.28	6654.00	5980.68
	Intel 6271C + T4	0.89	105.00	101.91	2149.31	1795.35	100	76.39	5176.00	4528.77
OCR-nopp-max640-mobile	Intel 6271C	1.00	1033.70	1005.95	2021.88	1743.27	N/A	N/A	N/A	N/A
	Intel 8350C	0.88	1043.60	1006.77	1980.82	1724.51	N/A	N/A	N/A	N/A
	Hygon 7490 + P800	0.28	125.70	101.56	1962.27	1782.68	N/A	N/A	N/A	N/A
	Intel 8350C + A100	0.21	122.50	101.87	1772.39	1569.55	29	18.74	2360.00	2039.07
	Intel 6271C + V100	0.43	133.80	101.82	1636.93	1464.10	37	20.94	2386.00	2055.30
	Intel 8563C + H20	0.18	119.90	102.12	2119.93	1889.49	29	20.92	2636.00	2321.11
	Intel 8350C + A10	0.24	126.80	101.78	1905.14	1739.93	48	30.71	2232.00	1911.18
	M4	7.08	137.80	104.83	2931.08	2658.25	N/A	N/A	N/A	N/A
	Intel 6271C + T4	0.36	124.80	101.70	1983.21	1729.43	61	46.10	2162.00	1836.63
OCR-nopp-max960-mobile	Intel 6271C	1.21	1020.00	1008.49	2200.30	1800.74	N/A	N/A	N/A	N/A
	Intel 8350C	1.01	1024.10	1007.32	2038.80	1800.05	N/A	N/A	N/A	N/A
	Hygon 7490 + P800	0.32	107.50	102.00	2001.21	1799.01	N/A	N/A	N/A	N/A
	Intel 8350C + A100	0.23	107.70	102.33	1727.89	1490.18	30	20.19	2646.00	2385.40
	Intel 6271C + V100	0.49	109.90	102.26	1726.01	1504.90	38	20.11	2498.00	2227.73
	Intel 8563C + H20	0.20	109.90	102.52	1959.46	1798.35	28	19.38	2712.00	2450.10
	Intel 8350C + A10	0.27	102.90	101.19	1938.48	1741.19	47	29.27	3344.00	2585.02
	M4	5.44	122.10	105.91	3094.72	2686.52	N/A	N/A	N/A	N/A
	Intel 6271C + T4	0.41	106.00	101.81	1859.88	1722.62	68	47.05	2264.00	2001.07
OCR-nopp-max640-server	Intel 6271C	2.16	1026.30	1005.10	3467.93	3074.06	N/A	N/A	N/A	N/A
	Intel 8350C	2.30	1008.70	1003.32	3435.54	3042.62	N/A	N/A	N/A	N/A
	Hygon 7490 + P800	0.35	104.70	101.27	1948.85	1779.77	N/A	N/A	N/A	N/A
	Intel 8350C + A100	0.25	104.90	101.42	1833.93	1560.71	41	27.61	4480.00	3955.14
	Intel 6271C + V100	0.56	106.20	101.47	1669.73	1500.87	58	31.78	3160.00	2838.78
	Intel 8563C + H20	0.23	109.40	101.45	1968.77	1800.81	58	30.81	2602.00	2588.77
	Intel 8350C + A10	0.30	106.10	101.55	2027.13	1749.07	69	39.10	3318.00	2795.54
	M4	7.26	133.90	104.48	5473.38	3472.28	N/A	N/A	N/A	N/A
	Intel 6271C + T4	0.58	103.90	100.86	1884.23	1714.48	84	63.50	2852.00	2540.37
OCR-nopp-max960-server	Intel 6271C	2.53	1014.50	1005.22	3625.57	3151.73	N/A	N/A	N/A	N/A
	Intel 8350C	2.66	1010.60	1003.39	3580.64	3197.09	N/A	N/A	N/A	N/A
	Hygon 7490 + P800	0.40	105.90	101.76	2040.65	1810.97	N/A	N/A	N/A	N/A
	Intel 8350C + A100	0.29	108.90	102.12	1821.03	1620.02	44	30.38	4290.00	2928.79
	Intel 6271C + V100	0.60	109.90	101.98	1797.75	1544.96	61	32.48	2936.00	2117.71
	Intel 8563C + H20	0.28	108.80	101.92	2016.22	1811.74	73	41.82	2636.00	2241.23
	Intel 8350C + A10	0.34	111.00	102.75	1964.21	1750.21	68	41.25	2722.00	2293.74
	M4	6.28	129.10	103.74	7780.70	3571.92	N/A	N/A	N/A	N/A
	Intel 6271C + T4	0.67	116.90	101.33	1941.09	1693.39	88	65.48	2714.00	1923.06
OCR-nopp-min1280-server	Intel 6271C	4.13	1043.40	1005.45	5993.70	3454.00	N/A	N/A	N/A	N/A
	Intel 8350C	4.46	1011.70	996.72	5633.51	3489.79	N/A	N/A	N/A	N/A
	Intel 8350C + A100	0.42	113.90	106.08	1747.88	1546.18	85	43.73	13558.00	11297.98
	Intel 6271C + V100	0.82	116.80	105.18	1873.38	1609.55	100	39.57	10376.00	8427.30
	Intel 8563C + H20	0.55	114.80	103.14	2036.36	1864.45	100	69.67	13224.00	11411.31
	Intel 8350C + A10	0.55	105.90	101.86	1931.35	1764.44	100	56.16	12418.00	10510.77
	Intel 6271C + T4	1.13	105.90	102.35	2066.73	1787.78	100	83.50	10142.00	8338.80
OCR-nopp-min1280-mobile	Intel 6271C	1.59	1019.90	1008.39	2366.86	1992.03	N/A	N/A	N/A	N/A
	Intel 8350C	1.29	1017.70	1007.28	2501.24	2059.99	N/A	N/A	N/A	N/A
	Hygon 7490 + P800	0.43	120.90	107.02	2108.87	1821.91	N/A	N/A	N/A	N/A
	Intel 8350C + A100	0.29	117.90	107.19	1847.97	1570.89	31	18.98	3746.00	3321.86
	Intel 6271C + V100	0.61	122.80	107.07	1789.25	1542.56	39	20.52	4058.00	3487.46
	Intel 8563C + H20	0.24	116.80	106.80	2092.63	1882.77	28	18.67	3902.00	3444.00
	Intel 8350C + A10	0.34	125.80	106.79	1959.45	1783.97	49	32.66	3532.00	3094.29
	M4	6.64	139.40	107.63	4283.97	3112.59	N/A	N/A	N/A	N/A
	Intel 6271C + T4	0.51	116.90	105.06	1927.22	1675.34	68	45.78	3828.00	3283.78

Pipeline configuration	description
OCR-default	Default configuration
OCR-nopp-mobile	Based on the default configuration, document image preprocessing is disabled and mobile det and rec models are used
OCR-nopp-server	Based on the default configuration, document image preprocessing is disabled
OCR-nopp-min736-mobile	Based on the default configuration, document image preprocessing is disabled, det model input resizing strategy is set to min+736, and mobile det and rec models are used
OCR-nopp-min736-server	Based on the default configuration, document image preprocessing is disabled, and the det model input resizing strategy is set to min+736
OCR-nopp-max640-mobile	Based on the default configuration, document image preprocessing is disabled, det model input resizing strategy is set to max+640, and mobile det and rec models are used
OCR-nopp-max960-mobile	Based on the default configuration, document image preprocessing is disabled, det model input resizing strategy is set to max+960, and mobile det and rec models are used
OCR-nopp-max640-server	Based on the default configuration, document image preprocessing is disabled, and the det model input resizing strategy is set to max+640
OCR-nopp-max960-server	Based on the default configuration, document image preprocessing is disabled, and the det model input resizing strategy is set to max+960
OCR-nopp-min1280-server	Based on the default configuration, document image preprocessing is disabled, and the det model input resizing strategy is set to min+1280
OCR-nopp-min1280-mobile	Based on the default configuration, document image preprocessing is disabled, det model input resizing strategy is set to min+1280, and mobile det and rec models are used

• Test environment:
  • PaddlePaddle 3.1.0、CUDA 11.8、cuDNN 8.9
  • PaddleX @ develop (f1eb28e23c)
  • Docker image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.1.0-gpu-cuda11.8-cudnn8.9
• Test data:
  • Test data containing 200 images from document and general scenarios.
• Test strategy:
  • Warm up with 20 samples, then repeat the full dataset once for performance testing.
• Note:
  • Since we did not collect device memory data for NPU and XPU, the corresponding entries in the table are marked as N/A.

CPU

Configuration	Avg. Time per Image (s)	Avg. Characters Predicted per Second	Avg. CPU Utilization (%)	Peak RAM Usage (MB)	Avg. RAM Usage (MB)
mobile_min_1280	2.00	326.44	976.83	2233.16	1867.94
mobile_min_736	1.75	371.82	965.89	2219.98	1830.97
mobile_max_960	1.49	422.62	969.11	2048.67	1677.82
mobile_max_640	1.31	459.11	978.41	2023.25	1616.42
server_min_1280	5.57	117.08	991.34	4452.39	3286.19
server_min_736	4.34	149.98	990.24	4020.85	3137.20
server_max_960	3.39	186.59	984.67	3492.62	2977.13
server_max_640	2.95	201.00	980.59	3342.38	2935.24

Deployment and Secondary Development

• Multiple System Support: Compatible with mainstream operating systems including Windows, Linux, and Mac.
• Multiple Hardware Support: Besides NVIDIA GPUs, it also supports inference and deployment on Intel CPU, Kunlun chips, Ascend, and other new hardware.
• High-Performance Inference Plugin: Recommended to combine with high-performance inference plugins to further improve inference speed. See High-Performance Inference Guide for details.
• Service Deployment: Supports highly stable service deployment solutions. See Service Deployment Guide for details.
• Secondary Development Capability: Supports custom dataset training, dictionary extension, and model fine-tuning. Example: To add Korean recognition, you can extend the dictionary and fine-tune the model, seamlessly integrating into existing pipelines. See Text Detection Module Usage Tutorial and Text Recognition Module Usage Tutorial for details.