dmmaze / comic-text-detector Goto Github PK

Hey I've been trying to get a command line tool which makes use of comic text detector to work, but I ran into some version incompatibilities I thought I should quickly mention.

On Numpy 1.23 your code seems to run just fine, however things break on the latest version 1.24 (which gets installed by default).
The problem seems to be that Numpy version 1.24+ removes the np.int alias.

Would be great if you could replace all uses of np.int with a suitable alternative? 😅
Not 100% sure but I think it's safe to replace np.int with np.int_ or just regular int (from my understanding of the problem)

how to fix it?

ModuleNotFoundError: No module named 'shapely'

File ~/imt/comic-text-detector/text_rendering.py:170, in TextLinesSampler.init(self, page_size, sampler_dict)
168 self.page_w, self.page_h = page_size
169 self.lang = sampler_dict['lang']
--> 170 self.lang_dict = load_dict(lang=self.lang)
171 self.orientation_sampler = ScaledSampler(sampler_dict['orientation'])
172 self.numlines_sampler = ScaledSampler(sampler_dict['num_lines'])

TypeError: load_dict() got an unexpected keyword argument 'lang'

In example.ipynb file, the code isn't work with error messages.
what can I do??

Static font files are missing for text rendering data/font_statics_en.csv and data/font_statics_jp.csv

I'm interested in leveraging features learned on Japanese manga to do transfer learning on the problem of detecting individual characters in a dataset of handwritten Japanese. While looking through the train_*.py scripts, I saw that you were providing a train_mask_dir argument. I realize that one of the outputs of comic-text-detector is an image mask so it makes sense that the model was trained on image segmentations annotations but I'm only interested in the text block detection module. How can I check out a model pretrained on comics and then fine-tune it with my dataset?

Auxiliary to this, you mention that “All models were trained on around 13 thousand anime & comic style images, 1/3 from Manga109-s” in the README. Does this mean that the entirety of Manga109-s was used during training and that the whole Manga109-s composed a third of the overall training dataset or that you only took one third of Manga-109s and then used this smaller subsample of Manga-109s in your overall training data? I'm wondering because Manga-109s does not provide image segmentation annotations. Did you just use the bounding box annotations or did you make use of the Manga109 image segmentation annotations made in the paper Unconstrained Text Detection in Manga?

ValueError                                Traceback (most recent call last)
Cell In[2], line 7
      5 img_dir = r'data/examples'                              # can be dir list
      6 save_dir = r'data/examples/annotations'
----> 7 model2annotations(model_path, img_dir, save_dir, save_json=False)

File ~/comic-text-detector/inference.py:45, in model2annotations(model_path, img_dir_list, save_dir, save_json)
     43     blk_xyxy.append(blk.xyxy)
     44     blk_dict_list.append(blk.to_dict())
---> 45 blk_xyxy = xyxy2yolo(blk_xyxy, im_w, im_h)
     46 if blk_xyxy is not None:
     47     cls_list = [1] * len(blk_xyxy)

File ~/comic-text-detector/utils/imgproc_utils.py:40, in xyxy2yolo(xyxy, w, h)
     39 def xyxy2yolo(xyxy, w: int, h: int):
---> 40     if xyxy == [] or xyxy == np.array([]) or len(xyxy) == 0:
     41         return None
     42     if isinstance(xyxy, list):

ValueError: operands could not be broadcast together with shapes (0,) (24,4) 

Thanks!

Can I make the instruction more detailed? Which file to run for training?

I would like to train a model in Spanish and I was wondering what dataset was needed? I mean the folder structure.