Hello, I am working on a research project, but my local computer doesnt have enough memory download the entore dataset, even in low quality. Is there a way i can download only the EPL data and labels of the EPL Data Per se. Would be a great help :)
Hi and sorry for the late reply. Unfortunately, I don't think there is a clean way to download only one game or one league at the moment with our pip package. An easy hack could be to start the download and hit CTRL+C (stopping the run of the python downloader) after the EPL games are downloaded (it should be the first games to download, for train, val and test separately). Not optimal, but it should work. Hope this helps, let us know.
This could be something we can look into if people are interested. For the most popular networks, I believe that there are already some really good videos made by other TH-camrs (for instance Yolo, U-NET etc.). Maybe what we could bring as value to the community are from-scratch implementations of our own action spotting (CALF) or real-time semantic segmentation (TinyNet) networks. Maybe taking the time to explain our choices of layers and so on in the process. Do you think that would be valuable?
What I have struggled with clean code implementation of computer Vision networks most TH-camrs only implement Unet because of its simple architecture as compared to other object detection algorithms and object tracking algorithms. It will be great to have a expert eye view on how to implement these to be familiar with best approaches when implementing computer Vision models from scratch and training models from scratch on big datasets and getting good results in production. I've been working on running computer Vision models on mobile devices training them from scratch and I have faced difficulties in training that's why I request you to consider my request it'll be great to have a mentor explaining and implementing computer Vision models from scratch.
I see. Their are indeed many difficulties when training a new model from scratch, and these difficulties also depend on the task that you're trying to achieve, the dataset on which you're training/evaluating and of course your network architecture. It would be very complex to cover all possible practical problems in a video series. What we could think of is a series of "good practice", or "stuff that should work" maybe. I will keep that in mind and discuss it with the team. Thank you for your suggestion.
Dear ANthony, There is some issue with 2_HQ 0r 1_HQ mkv files. Because even after giving the password they are showing unauthorized. While downloading the other mkv file I am able to download it with the password. Even in this video I can see first you wrote 1_mkv then all of a sudden you skipped it and wrote 1.mkv. SO please help me out man.
Hi, could you please try these lines and let me know if it works? pip install SoccerNet --upgrade python import SoccerNet from SoccerNet.Downloader import SoccerNetDownloader mySoccerNetDownloader=SoccerNetDownloader(LocalDirectory="path/to/SoccerNet") mySoccerNetDownloader.password = input("Password for videos (received after filling the NDA)") mySoccerNetDownloader.downloadGames(files=["1_720p.mkv", "2_720p.mkv"], split=["train","valid","test","challenge"])
Sorry about that, the video is a bit old and new names were given to the HQ videos for clarity reasons, maybe that's where the issue comes from. If that was actually the issue, we will see internally if we can re-activate somehow the old video name as a duplicate of the new one. In the meantime, please refer to our website to have the latest updates: www.soccer-net.org/download. Hope this helps.
Hello, I am working on a research project, but my local computer doesnt have enough memory download the entore dataset, even in low quality.
Is there a way i can download only the EPL data and labels of the EPL Data Per se.
Would be a great help :)
Hi and sorry for the late reply. Unfortunately, I don't think there is a clean way to download only one game or one league at the moment with our pip package. An easy hack could be to start the download and hit CTRL+C (stopping the run of the python downloader) after the EPL games are downloaded (it should be the first games to download, for train, val and test separately). Not optimal, but it should work. Hope this helps, let us know.
Can you do from scratch implementations of some computer Vision models using pytorch or tensorflow
This could be something we can look into if people are interested. For the most popular networks, I believe that there are already some really good videos made by other TH-camrs (for instance Yolo, U-NET etc.). Maybe what we could bring as value to the community are from-scratch implementations of our own action spotting (CALF) or real-time semantic segmentation (TinyNet) networks. Maybe taking the time to explain our choices of layers and so on in the process. Do you think that would be valuable?
What I have struggled with clean code implementation of computer Vision networks most TH-camrs only implement Unet because of its simple architecture as compared to other object detection algorithms and object tracking algorithms. It will be great to have a expert eye view on how to implement these to be familiar with best approaches when implementing computer Vision models from scratch and training models from scratch on big datasets and getting good results in production.
I've been working on running computer Vision models on mobile devices training them from scratch and I have faced difficulties in training that's why I request you to consider my request it'll be great to have a mentor explaining and implementing computer Vision models from scratch.
I see. Their are indeed many difficulties when training a new model from scratch, and these difficulties also depend on the task that you're trying to achieve, the dataset on which you're training/evaluating and of course your network architecture. It would be very complex to cover all possible practical problems in a video series. What we could think of is a series of "good practice", or "stuff that should work" maybe. I will keep that in mind and discuss it with the team. Thank you for your suggestion.
Thank you will wait for more awesome videos
Awesome, thanks!
Dear ANthony, There is some issue with 2_HQ 0r 1_HQ mkv files. Because even after giving the password they are showing unauthorized. While downloading the other mkv file I am able to download it with the password. Even in this video I can see first you wrote 1_mkv then all of a sudden you skipped it and wrote 1.mkv. SO please help me out man.
Hi, could you please try these lines and let me know if it works?
pip install SoccerNet --upgrade
python
import SoccerNet
from SoccerNet.Downloader import SoccerNetDownloader
mySoccerNetDownloader=SoccerNetDownloader(LocalDirectory="path/to/SoccerNet")
mySoccerNetDownloader.password = input("Password for videos (received after filling the NDA)")
mySoccerNetDownloader.downloadGames(files=["1_720p.mkv", "2_720p.mkv"], split=["train","valid","test","challenge"])
Sorry about that, the video is a bit old and new names were given to the HQ videos for clarity reasons, maybe that's where the issue comes from.
If that was actually the issue, we will see internally if we can re-activate somehow the old video name as a duplicate of the new one.
In the meantime, please refer to our website to have the latest updates: www.soccer-net.org/download. Hope this helps.