roman-vygon / triplet_loss_kws

Learning Efficient Representations for Keyword Spotting with Triplet Loss

triplet_loss_kws's Issues

Model not Learning

Hi!

First of all, thank you for open souring the code. I have tried to replicate the results and I have found a few issues during the training process.

1. I have generated a script following the presented notebook to generate dists.npy not present in the source code. The file is 799.9mb long that saves an array of shape (9998, 9998).
2. The classes probabilites files are missing, I am assigning them to None.
3. I had to comment the line in l2.py to avoid getting the grad_fn error while training.

After all of that, I can load your pretrained model Res15_35 (as there are no manifests files for 12 yet provided) and I can achieve the accuracy on Triplet evaluation. On the other hand though, there's no learning when training my model from scratch. The command used follows:

python TripletEncoder.py --name=test_encoder --manifest=35 --mode=Res15 --per_class=5 --per_batch=10 --hidden_size=45

Several per_batch and per_class parameters have been tested and same behaviour: The Triplet loss is always oscillating around 1.1 and 0.7 but there's not an evident decrease while training.

Then running the infer train script through:
python infer_train.py --name=res15_encoder --manifest=35 --model=Res15 --enc_step=25440 --hidden_size=45

The resulting Avg Accuracy is arround 20-35. This is not happening when loading the pretrained model, do you know what could be happening?

Thanks in advanced,

Biel.

Memory Issues

Hello Author,
I'm trying to train the TripletEncoder.py but GPU throws CUDA out of memory error. Can you specify the memory requirements to train this code?

Torch autograd expects Variable with requires_grad set to True

Torch autograd expects Variable with requires_grad set to True but doesn't find such Variables. Any idea as to how to get around this?

Package version mismatch while running with CUDA

Hi there,

Thank you for the great repo! While installing the packages listed in the Readme file using CUDA, I came across some version conflicts and circular dependencies in packages which are difficult to resolve. If possible, can anyone send through a requirements.txt file? It would make it easier and would be highly appreciated.

Kind Regards

How to generate the libri100.json?

Hi~
I follow your README.md
But I didn't find anywhere to get libri100_train.json, libri100_dev.json, libri100_test.json
to convert LibriWords manifests with convert_path_prefix.ipynb

KNN and Pretrained Models

Hi there,

Thank you for the great repo! Can I please know if you guys have also open sourced the KNN code you have used after triplet loss trained representations and the pre-trained models on triplet loss?

Thank You

Segmentaton Fault problem

After installing the required libraries. I have been trying to execute your pipeline. It is giving Segmentation fault (core dumped) every time. Can you pls share requirement.txt and also ideas to combat this problem?

Pre-trained Res8 models and parameters

Dear authors
With what parameters did you train Res8 with manifest 12?
Could you provide the pre-trained model? It's missing in the run folder.
Thank you for your help!

Google drive link invalid

Dear author, this google link seems invalid, can you just reshare the files 'convert_path_prefix.ipynb'? Thanks a lot!

Could the pretrained weight be provided?

Could the pretrained weight be provided?
thanks