Skip to content

Latest commit

 

History

History
71 lines (49 loc) · 1.8 KB

README_text.md

File metadata and controls

71 lines (49 loc) · 1.8 KB

Super Resolution for Scanned Text Images

The tiny model to upscale scanned images with text. The model uses the ConvTranspose2d layer instead of PixelShuffle, so the model can be launched on GPU and MYRIAD devices and Inference Engine support reshape function.

Train and Evaluation

Prepare Dataset

Create two directories for train and test images. Train images may have any resolution higher than the path_size. Validation images should have the resolution like the path_size.

./data
├── train
│   ├── 000000.png
│   ...
└── val
    ├── 000000.png
    ...

NOTE: Image should be in the gray scale format and contain only black (0) and white (255) pixels.

TIP: It is better to use cropped images like 500x500, because large resolution dramatically increases the time to read images.

Training

Use the tools/train.py script to start the training process:

python3 tools/train.py --config configs/text_scale3.yaml

To start from the pretrained checkpoint, set init_checkpoint in config.

Test

Use the tools/test.py script to evaluate the trained model:

python3 tools/test.py --test_data_path PATH_TO_TEST_DATA \
    --models_path PATH_TO_MODELS_PATH \
    --exp_name EXPERIMENT_NAME

Export to OpenVINO™

python3 tools/export.py --models_path PATH_TO_MODELS_PATH \
    --exp_name EXPERIMENT_NAME \
    --input_size 200 200 \
    --data_type FP32

Demo

For the Latest Checkpoint

python3 tools/text/infer.py --model PATH_TO_CHECKPOINT IMAGE_PATH

For Intermediate Representation (IR)

python3 tools/text/infer_ie.py --model <PATH_TO_IR_XML> \
    --device CPU \
    image_path