Comments (2)
@13331112522 Thanks for the questions, here are some answers:
How much impact it will have on the mIoU when I adjust size from 512 to 384 or 256 with --size. I got quite lower result compared with your result. The mIoU of the referit val down to 0.47 and unc to 0.36 when I use --size. I am wondering whether it is reasonable result.
The dropdown in performance is due to the evaluation process, in which the resulting segmentation map of DMN (which should be of size 256) is directly upsampled to the image original size (~600px) and then compared against the ground truth. Thus, details are lost, due to the reduction in the receptive field of the CNN neurons.
I am thinking of using bounding box rather than mask to locate the object with language guide. How could I use your code to train and test that? Is there any suggestions or instructions.
As segmentation maps contain more information than bounding boxes, you could take the resulting map and compute the enclosing bounding box. Otherwise, major modifications would be required on the original architecture to produce bounding boxes by default, such as removing the upsampling module, and take the Synthesis Module last hidden state to produce box coordinates, which are compared using Smooth-L1 loss in logarithmic space.
from dms.
Great thanks for your explanation and suggestion. @andfoy
from dms.
Related Issues (19)
- [Errno 2] No such file or directory: 'data/unc/corpus.pth' HOT 4
- ECCV objectives
- Getting very low mIoU using pre-trained weights for DMN HOT 6
- DataPrepare HOT 9
- I cannot not obtain your data. HOT 1
- Question of REFER HOT 6
- AttributeError: 'Program' object has no attribute '_program' HOT 2
- How much GPU does this work usually require? HOT 17
- The result problem HOT 4
- Weights Fine Tuning HOT 40
- Dissect ViLSTM
- Why net.train() in evaluate function? HOT 2
- How to visualize the output? HOT 2
- Testing on Custom Dataset HOT 24
- No such file or directory: 'data\\unc\\corpus.pth'? what is this corpus.pth? Need download? HOT 2
- Cannot produce results using the pretrained model HOT 3
- help me, can't download the referit_splits.tar.bz2 HOT 2
- Pretrained weights link seems to be down HOT 3
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 dms.