HWR challenge for index cards (only recognition, no detection)

Handwriting Recognition for Polish index cards. [ver. 2.0.0]

# submitter when ver. description dev-0 CER dev-0 WER test-A CER test-A WER
6 p/tlen 2022-09-06 06:23 2.0.0 Donut + MBart-50 large + mbart tokenizer donut mbart mbart-large-50 37.7±2.2 68.5±3.2 38.5±2.7 67.1±3.5
3 p/tlen 2022-09-06 05:02 2.0.0 Donut + MBart-50 large donut mbart mbart-large-50 16.5±1.7 49.5±3.5 19.3±2.0 54.9±4.2
2 p/tlen 2022-09-05 07:47 2.0.0 Donut donut 16.3±1.3 49.9±3.5 18.7±1.7 53.0±3.9
1 p/tlen 2022-05-17 05:04 2.0.0 TrOCR large finetune out-fixed batch-size=16 beam=10 fine-tuning=true lr=2.0e-5 max-epochs=300 model=trocr-large-handwritten patience=20 prediction-batch-size=16 [...] fine-tuned large 14.1±1.2 47.6±3.7 16.2±1.6 50.7±4.4
12 p/tlen 2022-05-17 04:58 2.0.0 Large fine-tuned N/A N/A N/A N/A
11 p/tlen 2022-05-17 04:55 2.0.0 Fix git annex mistake N/A N/A N/A N/A
10 p/tlen 2022-05-17 04:52 2.0.0 Fix git annex mistake N/A N/A N/A N/A
5 p/tlen 2022-05-17 04:51 2.0.0 TrOCR fine-tuned batch-size=24 beam=10 fine-tuning=true lr=2.0e-5 max-epochs=300 model=trocr-base-handwritten patience=20 prediction-batch-size=16 [...] fine-tuned base 19.2±1.7 54.6±3.5 20.9±2.0 55.4±4.8
4 p/tlen 2022-05-16 19:57 2.0.0 TrOCR fine-tuned batch-size=24 beam=10 fine-tuning=true lr=2.0e-5 max-epochs=300 model=trocr-base-handwritten patience=20 prediction-batch-size=16 [...] fine-tuned base 19.2±1.7 54.6±3.5 20.9±2.0 55.4±4.8
7 p/tlen 2022-05-16 17:19 2.0.0 TrOCR fine-tuned batch-size=8 beam=10 fine-tuning=true lr=2.0e-5 max-epochs=10 model=trocr-small-handwritten patience=2 prediction-batch-size=16 [...] fine-tuned small 75.9±2.8 136.3±5.5 75.0±2.9 141.9±6.1
8 p/tlen 2022-05-16 17:04 2.0.0 TrOCR fine-tuned beam=10 fine-tuning=false model=trocr-small-handwritten prediction-batch-size=16 small no-fine-tuning 145±25 175±20 118.0±9.3 160±11
9 p/tlen 2022-05-14 19:12 1.0.0 TrOCR fine-tuned batch-size=8 beam=10 lr=2.0e-5 max-epochs=10 model=trocr-small-handwritten patience=2 prediction-batch-size=16 validation-batch-size=16 [...] small 74.8±3.1 130.4±4.8 1521±192 761±41

Submission graph

Graphs by parameters

batch-size

[direct link]
[direct link]