| import numpy as np |
| from skimage import morphology |
|
|
| from annotator.teed import TEEDDetector |
| from annotator.util import HWC3 |
| from scripts.supported_preprocessor import Preprocessor, PreprocessorParameter |
| from scripts.utils import resize_image_with_pad |
|
|
|
|
| class PreprocessorTEED(Preprocessor): |
| def __init__(self): |
| super().__init__(name="softedge_teed") |
| self.tags = ["SoftEdge"] |
| self.slider_1 = PreprocessorParameter( |
| label="Safe Steps", |
| minimum=0, |
| maximum=10, |
| value=2, |
| step=1, |
| ) |
| self.model = None |
|
|
| def __call__( |
| self, |
| input_image, |
| resolution, |
| slider_1=None, |
| slider_2=None, |
| slider_3=None, |
| **kwargs |
| ): |
| img, remove_pad = resize_image_with_pad(input_image, resolution) |
| if self.model is None: |
| self.model = TEEDDetector() |
|
|
| result = self.model(img, safe_steps=int(slider_1)) |
| return remove_pad(result) |
|
|
|
|
| def get_intensity_mask(image_array, lower_bound, upper_bound): |
| mask = image_array[:, :, 0] |
| mask = np.where((mask >= lower_bound) & (mask <= upper_bound), mask, 0) |
| mask = np.expand_dims(mask, 2).repeat(3, axis=2) |
| return mask |
|
|
|
|
| def combine_layers(base_layer, top_layer): |
| mask = top_layer.astype(bool) |
| temp = 1 - (1 - top_layer) * (1 - base_layer) |
| result = base_layer * (~mask) + temp * mask |
| return result |
|
|
|
|
| class PreprocessorAnyline(Preprocessor): |
| def __init__(self): |
| super().__init__(name="softedge_anyline") |
| self.tags = ["SoftEdge"] |
| self.slider_resolution = PreprocessorParameter( |
| label="Resolution", |
| minimum=64, |
| maximum=2048, |
| value=1280, |
| step=8, |
| visible=True, |
| ) |
| self.slider_1 = PreprocessorParameter( |
| label="Safe Steps", |
| minimum=0, |
| maximum=10, |
| value=2, |
| step=1, |
| ) |
| self.preprocessor_deps = ["lineart_standard"] |
| self.model = None |
|
|
| def __call__( |
| self, |
| input_image, |
| resolution, |
| slider_1=None, |
| slider_2=None, |
| slider_3=None, |
| **kwargs |
| ): |
| img, remove_pad = resize_image_with_pad(input_image, resolution) |
| if self.model is None: |
| self.model = TEEDDetector(mteed=True) |
|
|
| mteed_result = self.model(img, safe_steps=int(slider_1)) |
| mteed_result = HWC3(mteed_result) |
| lineart_preprocessor = Preprocessor.get_preprocessor("lineart_standard") |
| assert lineart_preprocessor is not None |
| lineart_result = lineart_preprocessor(img, resolution) |
| lineart_result = get_intensity_mask( |
| lineart_result, lower_bound=0, upper_bound=1 |
| ) |
| cleaned = morphology.remove_small_objects( |
| lineart_result.astype(bool), min_size=36, connectivity=1 |
| ) |
| lineart_result = lineart_result * cleaned |
| final_result = combine_layers(mteed_result, lineart_result) |
| return remove_pad(final_result) |
|
|
|
|
| Preprocessor.add_supported_preprocessor(PreprocessorTEED()) |
| Preprocessor.add_supported_preprocessor(PreprocessorAnyline()) |
|
|
