Replicate study (that is, implementation) of 'DenseCap: Fully Convolutional Localization Networks for Dense Captioning' paper with tensorflow
Project is frozen. Only RPN part is implemented, but it's now working as intended. It was started as experiment but now it takes too much effort with no or little progress. Better to switch and (maybe) return to it later with more experience and knowledge. Current known problems are:
Converting sparse IndexedSlices to a dense Tensor of unknown shape. This may consume a large amount of memory
message somewhere inside_box_params_loss
fucntion.- Memory leak
- No Localization Layer
- No Language Layer
- Model not actually working ;)