Why weight model only have 2.5m ?

[ThorPham]

I training model with architect "Dense_unet " but when save params only 2.5m . When i find your params on baidu i see weight model very large 500m.
After 10000 step iou = [1.] mean_iou = [1.] i test on image but result very bad . Nothing segmentation. Can you help ?

[universea]

Sorry, Dense_unet have some errors and we have not enough time to fix it. In this competition, we used unet and get 0.54 score. The final score 0.60763 was promoted by post-processing.