search

hellochick / ICNet-tensorflow

TensorFlow-based implementation of "ICNet for Real-Time Semantic Segmentation on High-Resolution Images".
405 stars 153 forks source link

Inference on ADE20K, output is different than those in repo #77

Closed AliBuildsAI closed 5 years ago

AliBuildsAI commented 6 years ago

Hi.

I downloaded the checkpoint files you provided on Google drive for ADE20K dataset, and using the following command on indoor_1.jpg, indoor_2.jpg, and outdoor_2.jpg, I got the following results. python inference.py --img-path=./input/outdoor_2.png --dataset=ade20k --filter-scale=2 --model=others

The problem is that they are not as accurate as the images in the output folder in the repository. May I ask the checkpoint files in the Google drive folder are updated or not? Or is there possible that something else went wrong?

(I am able to replicate your result on CityScapes dataset.)

indoor_1 indoor_2 outdoor_2

hellochick commented 5 years ago

Hey,

I am not sure if the checkpoint is the same or not because there are several versions. I'm sorry for that. But I think the checkpoints might be only a little different between each others. I will update the whole repo in recent days!