amistele/MTARSI-fixed

MTARSI dataset, but actually correctly labeled

The Multi-type Aircraft of Remote Sensing Images (MTARSI) original dataset https://zenodo.org/records/3464319

• The original MTARSI dataset has a number of aircraft sorted into incorrect categories, and multiple categories completely mislabeled
• This fixes that issue, by re-defining labels based on the aicraft actually in the dataset, and ensuring all aircraft are actually in their correct categories
• Removed some images due to poor quality, e.g. photoshop errors in synthetic image generation that resulted in a commercial airline having only 3 engine nacelles
• Sample image grids are included to indicate how things have been resorted from original MTARSI, and illustrate a number of errors that existed in the original dataset
• As in the original MTARSI dataset, not all images are square. For modified versions of this dataset, see:
  • All images square: https://huggingface.co/datasets/amistele/MTARSI-fixed-square
  • All images 128x128: https://huggingface.co/datasets/amistele/MTARSI-fixed-128x128
  • All images 64x64: https://huggingface.co/datasets/amistele/MTARSI-fixed-64x64

Labels (This dataset, fixed MTARSI):

• A-10 : 340
• B-1 : 513
• B-2 : 619
• B-29 : 23
• B-52 : 548
• C-130 : 599
• C-135 : 654
• C-17 : 354
• C-5 : 631
• Commercial 2-engine : 546
• Commercial 4-engine : 47
• E-2 : 24
• E-3 : 430
• F-15 : 359
• F-16 : 391
• F-18 : 73
• F-22 : 42
• F-35 : 240
• F-4 : 29
• General Aviation : 434
• KC-10 : 368
• P-3 : 431
• Private Jet : 495
• Su-34 : 26
• T-43 : 306
• T-6 : 248
• Twin Prop : 374
• TOTAL: 9144

Original MTARSI labels (not actually included here):

• A-10 : 345
• A-26 : 230
• B-1 : 513
• B-2 : 619
• B-29 : 321
• B-52 : 548
• Boeing : 605
• C-130 : 763
• C-135 : 526
• C-17 : 480
• C-21 : 491
• C-5 : 499
• E-3 : 452
• F-16 : 372
• F-22 : 846
• KC-10 : 554
• P-63 : 305
• T-43 : 306
• T-6 : 248
• U-2 : 362
• TOTAL: 9385