[ad_1]
1.Run the pretrained pix2pix DNN (DNN1)
2.Implement and briefly describe two fine tuning strategies, i.e., one should be adding one-two final layers to DNN1 and train on the xBD2seg-train set to finetune DNN2 to segment xBD2seg-test set.
3.Implement the Intersection-over-union (IoU) metric to quantify ability of the network at segmenting floods from satellite images of size.
4.Report IoU for 10 different splits train/test of xBD2Seg dataset.
5.According to literature, an IoU of 0.5-0.65 is the state of the art on segmentation tasks over high resolution satellite imagery. Can fine tuning of DNN2 achieve this performance?
6.Compare your fine tuned network to the results you obtain from training a model from scratch using 256×256 cropos over our 1024×1024 xBD2seg dataset.
Sample Solution
The post Artificial Intelligence and Deep Learning appeared first on acestar tutors.
[ad_2]
Source link