experiments

Experiments

Applied a combination of RMSE, BCEwithLogitsLoss and Dice Loss on the DepthNet model. Result: Doesn't yeild good results!

Predictions

Segmentation Masks	Depth Map

Results

IOU	Validation Loss

Applied enhanced BCEwithLogitsLoss with Rmse Loss and Learning Rate=0.01. Result: Almost same results as that of DepthNet

Predictions

Segmentation Masks	Depth Map

Results

IOU	Validation Loss

Predictions

Applied Data Augmentation

• HueSaturationValue
• RandomBrightnessContrast Result: Predicted images are blurry! There is no necessity to use data augmentation transformations as the network is not over fitting.

Predictions

Segmentation Masks	Depth Map

Results

IOU	Validation Loss

|

Predictions

Created a architecture similar to U-NET, applied a combination of BCEWithLogitsLoss and SSIM Result: Results were similar to DepthNet, but since there are more than 3x parameters in U-NET, so I preferred DepthNet.

Predictions

Segmentation Masks	Depth Map

Results

IOU	Validation Loss