Closed imgreattt closed 2 years ago
Hello @dwipdalal,
first, thank you for your interest in our work! If you followed the instructions on the README, the needed files are generated automatically. In particular, scalars.csv
stores the training losses to enable the selection of a well-trained model.
Can you inform me about the steps, you already have run? Did you train the model, via the model
folder, and you want to evaluate it? Also, can you tell me the structure of your experimental setup? The subfolders ../SumMe/logs/split0/scalars.csv
seems right.
Yes, I ran the main.py file in the model folder. I have also run the inference.py file. This is the structure of the subfolders in my case is as follows:
The inference folder lets you evaluate our pretrained models either on the benchmarking datasets or on a given video.
If you have trained a model on a given dataset (as in your case), you need the evaluation folder. As mentioned in the README, you must change the appropriate PATHs to your structure. From the picture provided, I can see that the base_path
in evaluate_exp.sh has to be set to <path_until_here>/Summaries/PGL-SUM/
and the following bits will be added automatically to create the scalars.csv
file.
PS. I can see that you have stored in the same folder 3 different runs. Probably you'll need to retain only 1 and discard the rest .tfevents.
files.
Very sorry for the delay in replying. Thanks a lot for your response. I got a general idea I shall try to work on it tomorrow and will let you know the results.
Sorry for any inconvenience caused, and thanks a lot for your guidance.
Closing due to inactivity. Feel free to re-open the issue or create a new one, for furthermore questions and queries.
I ran the main.py file and then created a folder called summaries and stored logs and results in it. But then when I try running the choose_best_epoch.py I get an error saying: No such file or directory: PGL-SUM/Summaries/PGL-SUM/SumMe/logs/split0/scalars.csv
I am unable to understand what are scalar.csv here and how should I get this directory. Your help will be highly appreciated. Thank you.