deva-ocr-75m-research / metrics.json
voidash's picture
Release 74.84M Nepali Hindi English research OCR with source and native Paddle weights
5f70891 verified
Raw History Blame Contribute Delete
2.94 kB
{
"physical_neural_values_per_format": 74842752,
"metrics": {
"character": {
"images": 7820,
"sets": {
"hw_nep_dev": {
"images": 300,
"chars": 1524,
"edits": 198,
"exact": 196,
"failed": 0,
"empty": 0,
"capped": 0,
"CER": 0.12992125984251968
},
"hw_hindi_dev": {
"images": 2000,
"chars": 12054,
"edits": 974,
"exact": 1440,
"failed": 0,
"empty": 0,
"capped": 0,
"CER": 0.08080305292848847
},
"hw_en_dev": {
"images": 976,
"chars": 42074,
"edits": 2564,
"exact": 219,
"failed": 0,
"empty": 0,
"capped": 0,
"CER": 0.06094024813423967
},
"real_dev": {
"images": 2000,
"chars": 43430,
"edits": 4860,
"exact": 1618,
"failed": 0,
"empty": 0,
"capped": 0,
"CER": 0.11190421367718167
},
"synth_dev": {
"images": 800,
"chars": 10050,
"edits": 263,
"exact": 675,
"failed": 0,
"empty": 0,
"capped": 0,
"CER": 0.026169154228855723
},
"en_dev": {
"images": 1744,
"chars": 46033,
"edits": 6415,
"exact": 908,
"failed": 0,
"empty": 0,
"capped": 0,
"CER": 0.13935654856298743
}
},
"generation": {
"do_sample": false,
"num_beams": 10,
"max_new_tokens": 257,
"min_length": 0,
"length_penalty": 1.0,
"early_stopping": false,
"use_cache": true,
"suppress_tokens": [
0,
1,
3
]
},
"native_source_exact": true,
"verification_sha256": "7df9ebd369fd4e323039639f1ce9602f95193e0d6a42fc20cba37ece7f052e1b"
},
"english": {
"images": 976,
"sets": {
"hw_en_dev": {
"images": 976,
"chars": 42074,
"edits": 974,
"exact": 605,
"failed": 0,
"empty": 0,
"capped": 0,
"CER": 0.023149688643818035
}
},
"generation": {
"do_sample": false,
"num_beams": 10,
"max_new_tokens": 256,
"min_length": 0,
"length_penalty": 1.0,
"early_stopping": false,
"use_cache": true
},
"native_source_exact": true,
"verification_sha256": "daa40c7506e76966acaa17313a015c4b7badb904a15aebd9ab5a26371b6098ab"
}
},
"normalization": "NFC codepoint CER; all samples, failures, empties and caps retained",
"explicit_language_branch": true,
"automatic_routing": false,
"blind_unseen_writer_page_evaluation": false,
"handwriting_4percent_target_achieved": false
}