search

TimoSaemann / ENet

ENet: A Deep Neural Network Architecture for Real-Time Semantic Segmentation
577 stars 276 forks source link

benchmark result on val #61

Open youngwanLEE opened 6 years ago

youngwanLEE commented 6 years ago

Hi,

Could you let me know how to evaluate and benchmark on Cityscapes testset?

I can't figure out how to save the segmentation result by submission format.

Thanks in advance.

youngwanLEE commented 6 years ago

I evaluated your ENet-model which was downloaded on CITYSCAPES val set.

test resolution : 1024x2048 (original resolution)

classes IoU nIoU

road : 0.751 nan sidewalk : 0.537 nan building : 0.685 nan wall : 0.126 nan fence : 0.250 nan pole : 0.439 nan traffic light : 0.283 nan traffic sign : 0.433 nan vegetation : 0.773 nan terrain : 0.411 nan sky : 0.668 nan person : 0.491 0.411 rider : 0.338 0.291 car : 0.692 0.636 truck : 0.082 0.040 bus : 0.000 0.000 train : 0.000 0.000 motorcycle : 0.140 0.165 bicycle : 0.514 0.386

Score Average : 0.401 0.241

categories IoU nIoU

flat : 0.811 nan nature : 0.781 nan object : 0.450 nan sky : 0.668 nan construction : 0.690 nan human : 0.541 0.466 vehicle : 0.665 0.587

Score Average : 0.658 0.526

the results was not good as original paper results.

Did you evaluate your model?

andeyeluguo commented 4 years ago

how do you get the result?