rochelleni / dat Goto Github PK
View Code? Open in Web Editor NEWLicense: Other
dat's People
Stargazers
Watchers
dat's Issues
About decode.loss_seg
Checkpoints of Segformer pretrained on SHIFT discrete
Hi, thanks for your novel work. I have tried to train but unable to provide same results as your results (source). Since the performance of the source model is important to the whole continuous adaptation and to encourage a fair comparison, would you please provide the checkpoints of Segformer pretrained on the SHIFT discrete dataset?
About the code repository
We found that the code is empty, when will you upload the code again?
Where can I run your method?
Hi, thanks for your impressive work!
Btw, I wanna reimplement your method, so what .sh file can I use it for running your code?
It seems like there are only base.sh, tent.sh, cotta.sh
Some suggestions of Your Experiments
Hi, congratulations on your excellent work! I have read your paper carefully and has some suggestions on it.
In your paper, you compared with the results of CoTTA using the learning rate of 0.00006/8. From our experimental results, CoTTA can have similar results as yours at the first visiting if they use the same learning rate 3e-4:
CoTTA (lr=3e-4, without test-time augmentations):
Fog: 70.29 / Night: 44.21 / Rain: 66.12/ Snow : 61.66
Yours (lr=3e-4, with test-time augmentations):
Fog 71.7/ Night 44.4/ Rain 65.4/ Sow 62.9
Note that the results can have a +-0.2 of vibration due to the stochastic restoration. Also, the order of the data streaming can affect the results. Better to check whether you conducted a sorting on the filenames.
Following attached the log of my results on CoTTA:
start adapting fog
[>>>>>>>>>>>>>>>>>>>>>>>>>>>>>] 400/400, 0.7 task/s, elapsed: 565s, ETA: 0s
writing results to work_dirs/res.pkl
per class results:
+---------------+-------+-------+
| Class | IoU | Acc |
+---------------+-------+-------+
| road | 94.54 | 99.22 |
| sidewalk | 66.54 | 74.63 |
| building | 80.68 | 97.04 |
| wall | 58.03 | 71.08 |
| fence | 25.57 | 27.1 |
| pole | 47.33 | 56.82 |
| traffic light | 42.42 | 48.28 |
| traffic sign | 70.22 | 90.1 |
| vegetation | 87.19 | 91.66 |
| terrain | 72.02 | 88.8 |
| sky | 97.86 | 99.12 |
| person | 63.85 | 79.33 |
| rider | 67.44 | 86.42 |
| car | 87.74 | 95.59 |
| truck | 77.07 | 81.63 |
| bus | 93.26 | 98.02 |
| train | 86.21 | 96.21 |
| motorcycle | 54.43 | 64.7 |
| bicycle | 63.06 | 86.48 |
+---------------+-------+-------+
Summary:
+--------+-------+-------+------+
| Scope | mIoU | mAcc | aAcc |
+--------+-------+-------+------+
| global | 70.29 | 80.64 | 94.0 |
+--------+-------+-------+------+
[>>>>>>>>>>>>>>>>>>>>>>>>>>>>>] 400/400, 0.7 task/s, elapsed: 565s, ETA: 0s
writing results to work_dirs/res.pkl
per class results:
+---------------+-------+-------+
| Class | IoU | Acc |
+---------------+-------+-------+
| road | 89.33 | 98.88 |
| sidewalk | 53.88 | 59.51 |
| building | 70.42 | 87.84 |
| wall | 35.28 | 47.38 |
| fence | 24.5 | 27.7 |
| pole | 48.73 | 61.75 |
| traffic light | 44.78 | 60.42 |
| traffic sign | 46.07 | 52.79 |
| vegetation | 40.09 | 87.76 |
| terrain | 27.55 | 65.01 |
| sky | 0.98 | 0.98 |
| person | 51.57 | 76.0 |
| rider | 42.65 | 52.77 |
| car | 78.23 | 90.77 |
| truck | 19.75 | 70.78 |
| bus | 44.25 | 46.19 |
| train | 58.46 | 60.9 |
| motorcycle | 26.96 | 30.42 |
| bicycle | 36.54 | 42.75 |
+---------------+-------+-------+
Summary:
+--------+-------+-------+-------+
| Scope | mIoU | mAcc | aAcc |
+--------+-------+-------+-------+
| global | 44.21 | 58.98 | 72.73 |
+--------+-------+-------+-------+
start adapting rain
[>>>>>>>>>>>>>>>>>>>>>>>>>>>>>] 400/400, 0.7 task/s, elapsed: 565s, ETA: 0s
writing results to work_dirs/res.pkl
per class results:
+---------------+-------+-------+
| Class | IoU | Acc |
+---------------+-------+-------+
| road | 86.59 | 97.76 |
| sidewalk | 57.17 | 62.53 |
| building | 90.86 | 97.03 |
| wall | 45.49 | 59.4 |
| fence | 33.19 | 38.46 |
| pole | 57.03 | 70.51 |
| traffic light | 67.61 | 76.57 |
| traffic sign | 66.47 | 81.57 |
| vegetation | 90.83 | 94.92 |
| terrain | 60.64 | 86.16 |
| sky | 97.91 | 98.45 |
| person | 53.54 | 71.4 |
| rider | 61.56 | 81.05 |
| car | 84.12 | 92.13 |
| truck | 41.19 | 60.1 |
| bus | 75.31 | 79.51 |
| train | 73.7 | 85.37 |
| motorcycle | 52.68 | 66.03 |
| bicycle | 60.39 | 75.82 |
+---------------+-------+-------+
Summary:
+--------+-------+-------+-------+
| Scope | mIoU | mAcc | aAcc |
+--------+-------+-------+-------+
| global | 66.12 | 77.62 | 93.07 |
+--------+-------+-------+-------+
start adapting snow
[>>>>>>>>>>>>>>>>>>>>>>>>>>>>>] 400/400, 0.7 task/s, elapsed: 566s, ETA: 0s
writing results to work_dirs/res.pkl
per class results:
+---------------+-------+-------+
| Class | IoU | Acc |
+---------------+-------+-------+
| road | 84.02 | 97.99 |
| sidewalk | 51.14 | 62.09 |
| building | 87.17 | 97.43 |
| wall | 49.85 | 62.4 |
| fence | 42.46 | 44.89 |
| pole | 60.95 | 72.05 |
| traffic light | 74.03 | 82.8 |
| traffic sign | 69.34 | 82.8 |
| vegetation | 83.19 | 87.17 |
| terrain | 4.4 | 4.99 |
| sky | 97.68 | 98.73 |
| person | 66.38 | 83.22 |
| rider | 43.74 | 80.12 |
| car | 85.6 | 95.38 |
| truck | 49.63 | 57.19 |
| bus | 59.0 | 69.45 |
| train | 85.67 | 88.14 |
| motorcycle | 23.89 | 28.61 |
| bicycle | 53.49 | 69.88 |
+---------------+-------+-------+
Summary:
+--------+-------+-------+-------+
| Scope | mIoU | mAcc | aAcc |
+--------+-------+-------+-------+
| global | 61.66 | 71.86 | 91.21 |
+--------+-------+-------+-------+
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.