Comments (9)
This is what I get with the xception65_cityscapes_trainfine checkpoint on cityscapes val split. The results are slightly lower than official TF model (80,42%). I suspect this difference to be related to issue #10 .
The evaluation was done with the official Cityscapes evaluation script.
from pytorch-segmentation.
@remtav Thanks. Your experiment result is obtained in cityscapes val dataset?
from pytorch-segmentation.
@yazhe2017 Yes! I edited my post accordingly. Thanks!
from pytorch-segmentation.
hi! Can you tell me how to use the official Cityscapes evaluation script to eval this model?Thanks a lot for your help!
from pytorch-segmentation.
This is what I get with the xception65_cityscapes_trainfine checkpoint on cityscapes val split. The results are slightly lower than official TF model (80,42%). I suspect this difference to be related to issue #10 .
The evaluation was done with the official Cityscapes evaluation script.
Why is there Nan and what is nIou?
from pytorch-segmentation.
is there Nan and what is nIou?
As can be seen in the details for each submission to cityscapes-dataset.com (ex.: https://www.cityscapes-dataset.com/method-details/?submissionID=1366), nIoU is instance IoU and only gives a result for categories with instance annotation (ex.: person 1, person 2, person 3, etc.).
from pytorch-segmentation.
from pytorch-segmentation.
Sorry for late reply.
I fixed issue #10 and added code to evaluate cityscapes.
The results are as follows:
classes IoU nIoU
--------------------------------
road : 0.984 nan
sidewalk : 0.866 nan
building : 0.931 nan
wall : 0.626 nan
fence : 0.635 nan
pole : 0.668 nan
traffic light : 0.698 nan
traffic sign : 0.800 nan
vegetation : 0.929 nan
terrain : 0.651 nan
sky : 0.954 nan
person : 0.832 0.645
rider : 0.644 0.452
car : 0.956 0.887
truck : 0.869 0.420
bus : 0.906 0.657
train : 0.834 0.555
motorcycle : 0.674 0.404
bicycle : 0.783 0.605
--------------------------------
Score Average : 0.802 0.578
--------------------------------
categories IoU nIoU
--------------------------------
flat : 0.988 nan
construction : 0.937 nan
object : 0.729 nan
nature : 0.931 nan
sky : 0.954 nan
human : 0.842 0.667
vehicle : 0.944 0.859
--------------------------------
Score Average : 0.904 0.763
--------------------------------
from pytorch-segmentation.
thanks a lot for you ! I'll try it again.
from pytorch-segmentation.
Related Issues (20)
- prediction failure on Cityscapes images HOT 6
- freeze BN on single GPU
- How to train custom dataset HOT 1
- TTA Method in DeeplabV3+ | Evaluation HOT 1
- can not convert mobilenetv2 HOT 4
- ModuleNotFound error pretrainedmodels after pip install
- I think there is a wrong in Focalloss.py
- Question about prediction
- ERROR:ZeroDivisionError: division by zero HOT 1
- Cityscapes classes IoU 0.802 train from scratch or use pretrained model?
- Training on FP16 HOT 1
- Discrepancy in performance compared to paper HOT 1
- Normalization to [-1, 1] for DeepLab
- How to serialize the model for libtorch?
- num_samples=0
- voc image not normalize HOT 1
- ImportError at converting the pretrained model
- Why most encoders dissapeared ? HOT 1
- How's the training result? HOT 2
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from pytorch-segmentation.