LinXueyuanStdio / LaTeX_OCR_PRO

:art: 数学公式识别增强版:中英文手写印刷公式、支持初级符号推导(数据结构基于 LaTeX 抽象语法树)Math Formula OCR Pro, supports handwrite, Chinese-mixed formulas and simple symbol reasoning (based on LaTeX AST).
GNU General Public License v3.0
1.11k stars 235 forks source link

预测结果全是 "{{{{{{{{{{{{{{{{{{{{{{" #26

Closed 1667 closed 4 years ago

1667 commented 4 years ago

2020-07-27 14:30:36,244:INFO: Epoch 334/430 2020-07-27 14:30:39,214:INFO: - Training: - 2s - loss: 45.024583 - perplexity: 66662036885159257554076190113792.000000 - lr: 0.000242 2020-07-27 14:30:39,214:INFO: - Config: (before evaluate, we need to see config) 2020-07-27 14:30:39,215:INFO: { "export_name": "data.json", "dir_images_train": "data/small/images/train/", "dir_images_test": "data/small/images/test/", "dir_images_val": "data/small/images/val/", "path_matching_train": "data/small/matching/train.matching.txt", "path_matching_val": "data/small/matching/test.matching.txt", "path_matching_test": "data/small/matching/val.matching.txt", "path_formulas_train": "data/small/formulas/train.formulas.norm.txt", "path_formulas_test": "data/small/formulas/test.formulas.norm.txt", "path_formulas_val": "data/small/formulas/val.formulas.norm.txt", "max_iter": null, "max_length_formula": 150, "bucket_train": true, "bucket_val": true, "bucket_test": true, "buckets": [ [ 240, 100 ], [ 320, 80 ], [ 400, 80 ], [ 400, 100 ], [ 480, 80 ], [ 480, 100 ], [ 560, 80 ], [ 560, 100 ], [ 640, 80 ], [ 640, 100 ], [ 720, 80 ], [ 720, 100 ], [ 720, 120 ], [ 720, 200 ], [ 800, 100 ], [ 800, 320 ], [ 1000, 200 ], [ 1000, 400 ], [ 1200, 200 ], [ 1600, 200 ], [ 1600, 1600 ] ] } 2020-07-27 14:30:39,216:INFO: { "export_name": "vocab.json", "unk": "_UNK", "pad": "_PAD", "end": "_END", "path_vocab": "data/small/formulas/vocab.txt", "min_count_tok": 0 } 2020-07-27 14:30:39,216:INFO: { "export_name": "training.json", "device": "cuda:0", "criterion_method": "CrossEntropyLoss", "n_epochs": 430, "batch_size": 2, "dropout": 127, "clip": -1, "lr_method": "Adam", "lr_scheduler": "CosineAnnealingLR", "lr_init": 0.1, "lr_min": 0.0001, "start_decay": 6, "end_decay": 13, "lr_warm": 0.001, "end_warm": 0 } 2020-07-27 14:30:39,216:INFO: { "export_name": "model.json", "model_name": "Img2Seq", "encoder_cnn": "vanilla", "positional_embeddings": true, "attn_cell_config": { "cell_type": "lstm", "num_units": 512, "dim_e": 256, "dim_o": 512, "dim_embeddings": 80 }, "decoding": "beam_search", "beam_size": 2, "div_gamma": 1, "div_prob": 0, "max_length_formula": 160 } 2020-07-27 14:30:39,217:INFO: - Evaluating... 2020-07-27 14:30:40,705:INFO: - Eval: BLEU-4 is 0.57 || ExactMatchScore is 0.00 || EditDistance is 6.79 || perplexity is -197160514126517370880.00 2020-07-27 14:30:40,705:INFO: - New best score (-197160514126517370880.00)! 2020-07-27 14:30:40,705:INFO: - Saving model... 2020-07-27 14:30:41,301:INFO: - Saved model in results/small/model_weights/ 2020-07-27 14:30:41,301:INFO: - Elapsed time: 5.06, learning rate: 0.00014

求指点,谢谢