DetectionMetrics v2 #233
DetectionMetrics v2 #233
Conversation
|
I have modified the project webpage to have a new entrypoint, which clearly states the current status and acknowledges v1 appropriately. Also, the previous docs are untouched and will still be accessible in https://jderobot.github.io/DetectionMetrics/v1 (not deployed yet). Here is an overview of the new docs' appearance: detection_metrics_docs.mp4The new automatically generated sphinx docs for the Python code are stored in I have also updated the README accordingly: |
sergiopaniego
left a comment
sergiopaniego
left a comment
There was a problem hiding this comment.
Huge improvement!!
I've left some small comments 😄
README.md
Outdated
|
|
||
| ### Project webpage [here](https://jderobot.github.io/DetectionMetrics) | ||
|
|
||
| *DetectionMetrics* is a family of toolkits designed to unify and streamline the evaluation of perception models across different frameworks and datasets. With the release of ***DetectionMetrics v1*** we introduced a versatile suite of tools focused on object detection, supporting cross-framework evaluation and analysis. [Cite our work](#cite) if you use it in your research! |
There was a problem hiding this comment.
Nice! We could move all the information related to v1 to the bottom of the README, with a clear reference or link at the beginning for easy navigation.
Since the paper references DetectionMetrics, we should make sure to include a mention of it in the README.
There was a problem hiding this comment.
Done. Can you check the new layout?
|
|
||
| # DetectionMetrics | ||
|
|
There was a problem hiding this comment.
We could include an architecture diagram and a video of the application, as we had before. I think that's really useful for users.
There was a problem hiding this comment.
I've added a simple diagram showcasing the classes included in the library. As soon as I have some GUI ready I'll post a video!
There was a problem hiding this comment.
Just upgraded it
There was a problem hiding this comment.
This diagram looks awesome!!
README.md
Outdated
| <tr> | ||
| <th><i>DetectionMetrics v2</i></th> | ||
| <th>💻 <a href="https://github.com/JdeRobot/DetectionMetrics">Code</a></th> | ||
| <th>🔧 <a href="/v2/installation">Installation</a></th> |
There was a problem hiding this comment.
I guess all this reference would work once we merge it
There was a problem hiding this comment.
my bad, relative paths won't work in README 😅 I've fixed it but it won't work until new pages are deployed
|
|
||
| ### Common | ||
| Install your deep learning framework of preference in your environment. We have tested: |
There was a problem hiding this comment.
Is the application multiplatform?
There was a problem hiding this comment.
in the sense of different OSs?
There was a problem hiding this comment.
yep, it was an issue with the first version due to ROS constraints but not anymore as I can see!
detectionmetrics/models/torch.py
Outdated
| from PIL import Image | ||
| import torch | ||
| from torch.utils.data import DataLoader, Dataset | ||
| from torchvision.transforms import v2 |
There was a problem hiding this comment.
Reading the code below, I find this alias v2 a little strange.
There was a problem hiding this comment.
Fixed, we now import v2 as transforms, which is more self-explanatory and resembles classic torchvision usage
|
|
||
| # DetectionMetrics |
There was a problem hiding this comment.
Since we're generating the new version, maybe we could move the docs/gh-pages to a new branch gh-pages to separate the docs from the actual application code. What do you think?
There was a problem hiding this comment.
I like the proposal. I'd try to merge it first and see if the current deployment works tho, because I have added plain HTML docs generated with Sphinx (docs/py_docs) and I'd like to check whether it's working or not (locally it does as shown in the video above)
sergiopaniego
left a comment
sergiopaniego
left a comment
There was a problem hiding this comment.
LGTM! You can merge it 😄
Let's merge this version and start improving it instead of continuing working on this already big branch 🚀
|
|
||
| ### Common | ||
| Install your deep learning framework of preference in your environment. We have tested: |
There was a problem hiding this comment.
yep, it was an issue with the first version due to ROS constraints but not anymore as I can see!
|
|
||
| # DetectionMetrics | ||
|
|
There was a problem hiding this comment.
This diagram looks awesome!!
A complete rewriting of DetectionMetrics focused on semantic segmentation (for now). The new approach is pure Python and provides classes that enable transparent access to different datasets (Rellis3D, GOOSE, our GAIA format) and tools for evaluation with different frameworks (PyTorch, Tensorflow).
More details are provided in the new
READMEfile.Tech stack
Pending infra
detectionmetricspackage to PyPIPending functionality:
Things to decide:
https://jderobot.github.io/DetectionMetrics/v1)