|about 5 hours ago||4 days ago|
|Apache License 2.0||Apache License 2.0|
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
[D] Colab TPU low performance
2 projects | reddit.com/r/MachineLearning | 18 Nov 2021
I wanted to make a quick performance comparison between the GPU (Tesla K80) and TPU (v2-8) available in Google Colab with PyTorch. To do so quickly, I used an MNIST example from pytorch-lightning that trains a simple CNN.
[D] How to avoid CPU bottlenecking in PyTorch - training slowed by augmentations and data loading?
2 projects | reddit.com/r/MachineLearning | 10 Nov 2021
We've noticed GPU 0 on our 3 GPU system is sometimes idle (which would explain performance differences). However its unclear to us why that may be. Similar to this issue
[P] An introduction to PyKale https://github.com/pykale/pykale, a PyTorch library that provides a unified pipeline-based API for knowledge-aware multimodal learning and transfer learning on graphs, images, texts, and videos to accelerate interdisciplinary research. Welcome feedback/contribution!
2 projects | reddit.com/r/MachineLearning | 25 Apr 2021
If you want a good example for reference, take a look at Pytorch Lightning's readme (https://github.com/PyTorchLightning/pytorch-lightning) It answers the 3 questions of "what is this", "why should I care", and "how do i use it" almost instantly
2 projects | reddit.com/r/pytorch | 24 Apr 2021
[D] Advanced Takeaways from fast.ai book
2 projects | reddit.com/r/MachineLearning | 23 Mar 2021
Lower precision training can help and on pytorch lightning is just a simple flag you can set
[D] How to be more productive while doing Deep Learning experiments?
10 projects | reddit.com/r/MachineLearning | 25 Feb 2021
First of all, use high-level ML frameworks (AllenNLP, PyTorch-Lightning). No need to write boilerplate code and implement standard ML approaches from scratch. Here are some suggestions (thought more NLP-focused) that I feel improved my research coding experience a lot.
DDP with model parallelism with multi host multi GPU system
1 project | reddit.com/r/pytorch | 7 Feb 2021
PyTorch Lightning Flash appears to be copying fastai (without any credit) [D]
2 projects | reddit.com/r/MachineLearning | 5 Feb 2021
According to the README it's patent pending, but I learned about that from this HN thread. Funny thing is I didn't even remember there was a snafu about patents, but looked it up because of some vague recollection of the PL founder getting into a tussle about some other trivial topic (apparently it was how well PyTorch works on TPUs).
[D] Training 10x Larger Models and Accelerating Training with ZeRO-Offloading
3 projects | reddit.com/r/MachineLearning | 25 Jan 2021
I also asked for the respective support in PytorchLightning in this issue: Add deepspeed support · Issue #817 · PyTorchLightning/pytorch-lightning (github.com)
Nicest and cleanest Deep Learning codebases out there
3 projects | reddit.com/r/deeplearning | 22 Jan 2021
When I look at the pytorch lightning animation, the stuff on the left for me is easy to follow and the code on the right formatted into classes is hard. My goal is to to start thinking and coding more like the code on the right. What I typically find hard with reading through code where everything is inside classes, methods, functions, decorators etc (i.e. the code on the right) is that there will be a place that executes all these methods in a linear way, but I keep having to scroll up to the class to see what it is actually doing. On the left I can just read through the code top to bottom. I even find myself copying the code out of classes the first time I read it so it executes like the code on the left :P I feel like what I'm doing is the equivalent of typing with only my index fingers…
[D] Good quality code repos on deep learning
6 projects | reddit.com/r/MachineLearning | 27 Aug 2021
[D] Mask R-CNN was from 2017 - are there any good newer instance segmentation models in the last 4 years that beat it? Does anyone know if D2Det (2020) outperforms it? What would you say is the best instance segmentation model out there right now? Thanks
1 project | reddit.com/r/MachineLearning | 21 Aug 2021
If you want to find implementations of new methods check out MMDet. You can find SCNet on there which is a new 2021 Cascaded (multi-stage) model. The last time I looked into newer detection models (a few months ago) D2Det was the best performing two-stage detector I could find and SCNet was the best cascaded model. In the SCNet paper their model appears to be about ~1 AP above D2Det using the same (ResNet 101) backbone
Master thesis on autonomous vehicles (cybersecurity aspect)
2 projects | reddit.com/r/AutonomousVehicles | 17 Jul 2021
You create and test the attacks on datasets like Kitti, NuScenes, and many others. Basically you try to manipulate the input to a certain detection pipeline for example (You can find a lot of LiDAR and camera based detection pipelines here: https://github.com/open-mmlab/mmdetection3d and here https://github.com/open-mmlab/mmdetection). You try to manipulate the input so that it deceives the car to do what you need without having control to the car itself.
[Project] Need help understanding code in MMDetection Library
1 project | reddit.com/r/MachineLearning | 5 Jul 2021
I look up into this MMDetection library to learn the basics, I read the manuals and tutorials. Now, I want to edit some parts of the NECK network (FPN), but the code in mmdetection/fpn.py at master · open-mmlab/mmdetection (github.com) doesn't have enough documentation/comments on what this part does, etc.
My first Pypi Package (make-voc-dataset)
3 projects | reddit.com/r/Python | 23 Jun 2021
Majority of the current Deep Learning Frameworks like MMDetection or Detectron2 support the VOC Formatted Data / COCO Formatted Data
[D] Are there any good 3rd party image segmentation/ object detection frameworks besides Tensorflow and Pytorch?
2 projects | reddit.com/r/MachineLearning | 16 Mar 2021
At least with Pytorch, it is trivial to Frankenstein things and just borrow whatever components you want from other repos. If you do want to look at a framework, I think MMDet is probably the only one that is actually used by some researchers to publish their own work: https://github.com/open-mmlab/mmdetection
Mask RCNN implementation in python
1 project | reddit.com/r/computervision | 11 Feb 2021
I’ve trained Mask RCNN in Google Colab using this Pytorch library - https://github.com/open-mmlab/mmdetection
What are some alternatives?
detectron2 - Detectron2 is FAIR's next-generation platform for object detection, segmentation and other visual recognition tasks.
yolov5 - YOLOv5 🚀 in PyTorch > ONNX > CoreML > TFLite
pytorch-grad-cam - Many Class Activation Map methods implemented in Pytorch for CNNs and Vision Transformers. Including Grad-CAM, Grad-CAM++, Score-CAM, Ablation-CAM and XGrad-CAM
Sacred - Sacred is a tool to help you configure, organize, log and reproduce experiments developed at IDSIA.
metaflow - :rocket: Build and manage real-life data science projects with ease!
yolact - A simple, fully convolutional model for real-time instance segmentation.
pytorch-forecasting - Time series forecasting with PyTorch
Mask_RCNN - Mask R-CNN for object detection and instance segmentation on Keras and TensorFlow
guildai - Experiment tracking, ML developer tools
tmux - tmux source code
sahi - A lightweight vision library for performing large scale object detection/ instance segmentation.