Anh Nguyen

The code used a quite old version of CUDA 7. You may want to setup cuda 7 or 8 to run the code. Newer cuda version is not tested.

Our core caffe code will work with multple GPUs (see [this](http://caffe.berkeleyvision.org/tutorial/interfaces.html)), however you may have to modify some IO utils to make them work with multi-gpus. Unfortunately, I don't have...

Currently, we use threshold=0.9, if no box > 0.9, we choose the highest one - no matter how big the confidence is. You can change the param `CONF_THRESHOLD = 0.9`...

Thanks for your interest @felixfuu! I just added the [utils](https://github.com/nqanh/affordance-net/tree/master/utils) folder. You can find the script to create .sm files and all relevant information to train AffordanceNet on your own...

The affordance mask depends on the size of the object, and the object size depends on the anchor parameters (scale and ratio) of the object detector. The concept of anchor...

No, we use VGG16 backbone to extract features and mainly focus on the mask branch (for multiclass affordances). The object detection branch is quite simple (only 2 fully connected layers...

I added One Drive links for all files. Hope you can access the data!

This is a numpy version problem. I guess you're using version > 1.11.2, which doesn't support float index. A quick fix is just to install numpy 1.11.2: `sudo pip install...

You can fix the float index problem by converting the indexes to integer. Or without root, you can try `pip install numpy==1.11.2 --user`. It will be installed on your local...

You can manually convert the float indexes to integers. If it works, then it's ok.