hkchengrex / vos-benchmark

Fast and general video object segmentation evaluation.
MIT License
24 stars 3 forks source link

Can we use it for other datasets rather than davis 2016? #3

Open JawadTawhidi opened 9 months ago

JawadTawhidi commented 9 months ago

Hi, can we use it for other data sets rather than davis 2016? my dataset is ultrasound videos and is similar to davis 2016, may I use this script for calculating J and F values?

If not, would you please give me some clues on how can I calculate the J and F values for my results?

JawadTawhidi commented 9 months ago

I used it to calculate J and F values on my ultrasound data set, during training process of my model when validation code is displaying the results after each epoch the highest J&F value was 66.737.

But when I used this repo for evaluation the J&F value is 66.4.

Now I am confused if this repo is working for evaluating my data set or not, would you please give me some clue?

One more thing I have to mention is that my data set has single object for most of the videos, but in some videos it has 3 to 4 objects as well.

hkchengrex commented 9 months ago

The script should work as long as the data is formatted as integer masks when read by PIL. Given that the difference in J&F is quite small, there might be slight implementation differences between your validation code and this repo. I do not know what those differences might be without knowing anything about your validation code.

hkchengrex commented 9 months ago

Hello. I cannot really look at all these code. I can help if there are specific questions regarding this repo.