2018 Data Science Bowl, Kaggle
2018 Data Science Bowl, Kaggle
posted in 2018
Embedded Files
(I) Objective: Identify all the cells’ nuclei in the image.
(II) Approach:
Model: Mask R-CNN with the NASNet feature extractor, instead of ResNet101 feature extractor.
non-maximum suppression: Using the IoU of the pixel area of the object, instead of the IoU of the bounding box area of the object.
Ensemble the test results: Ensemble the test results of 0, 90, 180, and 270 rotation images.
(III) Results: The score is 0.5 on the Kaggle Private Leaderboard, 147th of 3634, Top5%. The score is the result evaluated on the mean average precision at different intersection over union (IoU) thresholds from 0.5 to 0.95.
Reference: https://www.kaggle.com/c/data-science-bowl-2018
(IV) The nuclei results of the Mask R-CNN with the NASNet.
(test case 1) :
(Fig. 1a) input image
(Fig. 1b) the predicted result
(test case 2) :
(Fig. 2a) input image
(Fig. 2b) the predicted result
(test case 3) :
(Fig. 3a) input image
(Fig. 3b) the predicted result
(test case 4) :
(Fig. 4a) input image
(Fig. 4b) the predicted result
(test case 5) :
(Fig. 5a) input image
(Fig. 5b) the predicted result
Google Sites
Report abuse