Comments (8)
I also met this two errors when I tried to test the RUGD dataset with the pre-trained model:
KeyError: 'RUGDDataset_Group is not in the dataset registry'
So I changed the code of the config file 'ganav_rugd_6' at line 71, 116, 138 and line 162 from RUGDDataset_Group
to RUGDDataset_Group6
, and it works.
Another error is :
KeyError: "EncoderDecoder: 'OursHeadClassAttNew is not in the models registry'"
For this, I changed the code in the same file at line 20 from OursHeadClassAttNew
to OursHeadClassAtt
. Then the testing code is ready to run.
Hope it helps
from ganav-offroad.
Thanks for the catch! The config is only updated in the trained_model folder but not updated on the google drive. This has been fixed.
Regarding your Q1, I'm actually not sure why this happened. What is your current resolution on the test image? I think you might have to choose a good padding for your own dataset (not too disproportional), if you are training from scratch. For example, if you want to use trained RUGD model to test on your own image, you have to use the same padding size for inference ((300, 375) for RUGD). But you have training image and labels, you can change the padding in the training and testing pipeline that fit you data best.
from ganav-offroad.
Hi, Thanks for the quick reply!
My current resolution on the test image is 960*540. It's a small dataset so I trained it based on the pre-trained model of the RUGD dataset, I didn't change the padding during the training process (it's still (300,375)), should I change it in this case?
from ganav-offroad.
I don’t have a good answer to this since I have not used this size. You can try both and see which works for you best. If you have a small dataset, I assume it would not generalize well.
I’m closing this issue since it’s regarding assertion error. Please open another issue if you have more questions.
from ganav-offroad.
Yeah, I understand.
I tried to change the crop size in the code, but I always get the AssertionError when the crop size is not (300, 375), I'm not sure if it's because the 'RUGDDataset_Group6' dataset type only compatiable with (300, 375).
And May I ask how did you define the dataset type (line 2) in /base/dataset/rugd_group6.py, it seems that the mmcv will check if this string is in the library. I'm wondering if the type of dataset is related to the crop size.
from ganav-offroad.
You can search for “ RUGDDataset_Group6” globally to figure out how to define the dataset. You need to create your own py files in base and config folder.
The provided model weight only takes 300 by 375 patches. But if you train your own model, the model weight can be anything. Everything including the dataset is define by yourself, so you need to walk through the code, figure out the details and see how to define your own dataset. You should also check the assertion and see what it’s checking. Another good source is from the original mmseg GitHub(https://github.com/open-mmlab/mmsegmentation/blob/master/docs/en/tutorials/customize_datasets.md). The structure of our code is based on mmseg, but only support our method.
from ganav-offroad.
The link provided just to help you understand the structure of the config definition, but we also made some changes in our repo as well. As to the assertion, as long as the training patch size and testing patch size are the same, there should be no error. (If you train with crop size (x, y), the model is created with weights that fits (x,y) patches, so there would be no problem during training and testing. I have not personally test other size, but I started from rugd and generalize the crop size to rellis-3D, so there should be no problem.
from ganav-offroad.
Thank you so much! I will read the tutorial carefully.
It did no errors when I changed the image size to 688*550, I think the crop size need to be adjusted according to different datasets.
from ganav-offroad.
Related Issues (20)
- Check point file HOT 10
- about the training loss become nan HOT 6
- Pre-processing images HOT 2
- Enquiry regarding Performance of Training HOT 14
- Error in ONNX conversion HOT 10
- how to get segmentation cost map from segmentation results? HOT 1
- Conversion of RUGD and Rellis Datasets To Rugd6 Group & Training HOT 4
- Is the onnx file generated by pytorch2onnx.py quantized? HOT 1
- Welcome update to OpenMMLab 2.0 HOT 1
- Use pretrained ResNet50 model (backbone) on ImageNet.
- L1? L2? L3? HOT 2
- checkpoint error HOT 4
- Output image of the model HOT 5
- ModuleNotFoundError: No module named 'mmcv._ext' HOT 4
- Change of backbone HOT 2
- Training on GOOSE Dataset HOT 3
- rellis folder structure
- output quality HOT 2
- User Guide / Usage 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 ganav-offroad.