prof-lu-cewu / visual-relationship-detection Goto Github PK

Hi,

Thanks for the code. It is a great work!

But can you also provide the training code? Then we can train our own W and b in Eq. (2).
And is that possible to provide your pre-trained model for the object detection and predicates prediction model, so we can extract the 'objectDetRCNN.mat' and 'UnionCNNfeaPredicate.mat' for our own images.

Thanks!

I am unable to figure out where to download the images. Thank You in advance

Hello,

I'm confused about the evaluation of predicate detection.
The paper says

our input is an image and set of localized objects. The task is to predict a set of possible predicates between pairs of objects

But in predicate_detection.m, the predictions are made for each groundtruth relation. It takes a groundtruth (subject, object) pair as input and classifies the predicate.
If that is the case, the model has access to not only an image and set of localized objects, but also:

1. Whether or not a pair of objects have a relation
2. (If they do have certain relation) The order of the two objects (which one is the subject and which one is the object)

Could you please clarify what is the correct setup for the predicate detection task? Thanks.

Hello，your paper help me a lot and I want to build my own dataset. But I don't know how to generate the json file conveniently. Do you have any tools to generate them(like labelImg).
look forward to your reply.

Hi,

Thanks you for the effort of releasing the source code.
In top_recall_Relationship.m, when calculating num_pos_tuple, you used
length(gt_tuple_label{ii}). length gives the dimension of the longer axis, which results in an erroneous denominator when calculating recall score. Would you consider fix and use size(gt_tuple_label{ii}, 1) instead?

Hello,How train Union CNN Score? Can you release the training code?

Hi,

Can you share with us the data for evaluating the zero-shot learning and image retrieval as stated in the paper?

Best,