icdar2015.yaml 2.0 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869
  1. name: DBNet
  2. dataset:
  3. train:
  4. dataset:
  5. type: ICDAR2015Dataset # 数据集类型
  6. args:
  7. data_path: # 一个存放 img_path \t gt_path的文件
  8. - ''
  9. pre_processes: # 数据的预处理过程,包含augment和标签制作
  10. - type: IaaAugment # 使用imgaug进行变换
  11. args:
  12. - {'type':Fliplr, 'args':{'p':0.5}}
  13. - {'type': Affine, 'args':{'rotate':[-10,10]}}
  14. - {'type':Resize,'args':{'size':[0.5,3]}}
  15. - type: EastRandomCropData
  16. args:
  17. size: [640,640]
  18. max_tries: 50
  19. keep_ratio: true
  20. - type: MakeBorderMap
  21. args:
  22. shrink_ratio: 0.4
  23. thresh_min: 0.3
  24. thresh_max: 0.7
  25. - type: MakeShrinkMap
  26. args:
  27. shrink_ratio: 0.4
  28. min_text_size: 8
  29. transforms: # 对图片进行的变换方式
  30. - type: ToTensor
  31. args: {}
  32. - type: Normalize
  33. args:
  34. mean: [0.485, 0.456, 0.406]
  35. std: [0.229, 0.224, 0.225]
  36. img_mode: RGB
  37. filter_keys: [img_path,img_name,text_polys,texts,ignore_tags,shape] # 返回数据之前,从数据字典里删除的key
  38. ignore_tags: ['*', '###']
  39. loader:
  40. batch_size: 1
  41. shuffle: true
  42. num_workers: 0
  43. collate_fn: ''
  44. validate:
  45. dataset:
  46. type: ICDAR2015Dataset
  47. args:
  48. data_path:
  49. - ''
  50. pre_processes:
  51. - type: ResizeShortSize
  52. args:
  53. short_size: 736
  54. resize_text_polys: false
  55. transforms:
  56. - type: ToTensor
  57. args: {}
  58. - type: Normalize
  59. args:
  60. mean: [0.485, 0.456, 0.406]
  61. std: [0.229, 0.224, 0.225]
  62. img_mode: RGB
  63. filter_keys: []
  64. ignore_tags: ['*', '###']
  65. loader:
  66. batch_size: 1
  67. shuffle: true
  68. num_workers: 0
  69. collate_fn: ICDARCollectFN