carlesventura

Would it be possible to release the model that is used in the demo online? If this model is better than the ones released, it would be benefitial for the...

In that case, the reason could be that the model is not able to find any other relevant instance in the image and I would choose the first mask given...

The code is not ready for this scenario. However, you could process all frames until say the 20th frame (no new objects entering), save the predictions at 20th frame, and...

Could you explain with more detail the problem you have? Regards, Carles

If you check the code from eval_previous_mask.py, you will find that the masks for each instance are also saved as binary images in a folder named masks_sep_2assess. This folder should...

Have you tried by generating first the LMDB file with the following command? python dataset_lmdb_generator.py -dataset=youtube Regards, Carles

The ideal should be by including a parameter that allows to choose the image resolution at which you want to train the model. However, in the current implementation, you should...

There could be at least two reasons for the different results when retraining: 1. A new training, will see the images and the instances in a different order, and the...

Dear huanxihhh, The results from DRIU have to be computed or can be downloaded from the DRIU website project: http://www.vision.ee.ethz.ch/~cvlsegmentation/driu/ Best regards, Carles On Sun, 16 Sep 2018 at 04:46,...