HHeracles

The structure of data directory is: dataset ├── annotations │ ├── captions_train2014.json │ ├── captions_val2014.json │ ├── coco_dev_ids.npy │ ├── IIIT5k_3000 │ ├── coco_restval_ids.npy │ ├── coco_test_ids.npy │ └── coco_train_ids.npy...

The structure of data directory is: dataset ├── annotations │ ├── captions_train2014.json │ ├── captions_val2014.json │ ├── coco_dev_ids.npy │ ├── IIIT5k_3000 │ ├── coco_restval_ids.npy │ ├── coco_test_ids.npy │ └── coco_train_ids.npy...

About computing text detection indicators

2

Thanks for your work. But I have two questions I'd like to ask. First, how do you calculate the P, R, and F of text detection? Do you need to...

Thanks for your work。When I reproduce your experiment, using the default parameters you provide, pretrain is much slower thanfinetune. Only about 30,000~35,000 iterations of pre-training were completed in five days....

Can I get a single character box？

4

Hello @MhLiao ，We can get each character's mask in your program，but each character's mask only covers a part of the character. Now, I want to get the feature map for...

How to reproduce the experimental data in the paper?

6

Hello @MhLiao .I used the downloaded model “trained_model.pth”，and set the MIN_SIZE_TEST parameter to the corresponding value，1000 or 1440.However,.The data obtained on data set Total_text( MIN_SIZE_TEST=1440) are respectively "hmean": 0.831966(strong),0.779740(weak),0.733809(generic).But the...

Evaluation: Calculated!{"precision": 0, "recall": 0.0, "hmean": 0, "AP": 0}

4

Hello @MhLiao ，when I have run the command “cd evaluation/icdar2015/e2e/” and “python script.py”，I used the downloaded model “model_finetune.pth” and the icdar2015.But the output is ”Calculated!{"precision": 0, "recall": 0.0, "hmean": 0,...

Thank you for your excellent work. Baidu Drive disk link failed. Can you provide a new link