Hicham Bellafkir

DCH: low results

3

I trained DCH on COCO using the default parameters in val_train script, which sets the iter_num to 2000. The results too low compared to the paper: iter 33 prec 0.3673785145624646...

False sampling of data

1

hi, I just found out, that all images in the query list are also in the database list, which is not allowed for fair validation. thanks

AlexNet on COCO

1

can please share the train parameter for Alexnet on coco?

The crop size is set to be 224x224, which is not the same crop size as in HashNet (227x227). Is there a reason for that?

hi, looking to models/predict/coco/predict_parallel.py script, you guys used np.sign as the sign function, but np.sign for x>0 returns 1 and for x

hey, can the implementation of contrastive center loss work on a multi-label dataset such as coco?

Switching between torchaudio and DFT mode results in different results. To reproduce: ``` FFT_SIZE = 1024 SR = 16_000 HOP_SIZE = 512 MEL_BANDS = 64 x = torch.rand(1, SR) wn...

### Is your feature request related to a problem? Most state-of-the-art deep learning approaches for audio tagging use Mel spectrograms rather than linear spectrograms. ### Describe the solution you'd like....