dong-jinkim / denserelationalcaptioning Goto Github PK

Is the relational_captions.json your final data for training the model? Could you provide the codes for preprocessing the raw VG dataset and explain how to get the relational_captions.json file? Thanks.

Hello,

We need some help to understand what the below issue might mean. We are on Ubuntu 18.04, NVIDIA Driver 418, CUDA 10.1. We were able to install all dependencies except Torch7 from the specified links. For Torch7, we used a third party repository that made some changes to make it work with CUDA 10. After running the script

th run_model.lua -input_image 60.jpg

we get -

th run_model.lua -input_image 60.jpg -gpu -1 -use_cudnn -1
Warning: cudnn.convert does not work with nngraph yet. Ignoring nn.gModule1/1 processing image 60.jpg	
/home/ruddra/torch/install/bin/luajit: /home/ruddra/torch/install/share/lua/5.1/nn/Container.lua:67: 
In 4 module of nn.Sequential:
In 1 module of nn.Sequential:
In 2 module of nn.ParallelTable:
In 1 module of nn.Sequential:
/home/ruddra/torch/install/share/lua/5.1/nn/Linear.lua:66: invalid arguments: FloatTensor number FloatTensor number CudaTensor FloatTensor 
expected arguments: *FloatTensor~2D* [FloatTensor~2D] [float] FloatTensor~2D FloatTensor~2D | *FloatTensor~2D* float [FloatTensor~2D] float FloatTensor~2D FloatTensor~2D
stack traceback:
	[C]: in function 'addmm'
	/home/ruddra/torch/install/share/lua/5.1/nn/Linear.lua:66: in function </home/ruddra/torch/install/share/lua/5.1/nn/Linear.lua:53>
	[C]: in function 'xpcall'
	/home/ruddra/torch/install/share/lua/5.1/nn/Container.lua:63: in function 'rethrowErrors'
	/home/ruddra/torch/install/share/lua/5.1/nn/Sequential.lua:44: in function </home/ruddra/torch/install/share/lua/5.1/nn/Sequential.lua:41>
	[C]: in function 'xpcall'
	/home/ruddra/torch/install/share/lua/5.1/nn/Container.lua:63: in function 'rethrowErrors'
	.../ruddra/torch/install/share/lua/5.1/nn/ParallelTable.lua:12: in function <.../ruddra/torch/install/share/lua/5.1/nn/ParallelTable.lua:10>
	[C]: in function 'xpcall'
	/home/ruddra/torch/install/share/lua/5.1/nn/Container.lua:63: in function 'rethrowErrors'
	...
	[C]: in function 'xpcall'
	/home/ruddra/torch/install/share/lua/5.1/nn/Container.lua:63: in function 'rethrowErrors'
	/home/ruddra/torch/install/share/lua/5.1/nn/Sequential.lua:44: in function 'forward'
	./densecap/DenseCapModel.lua:237: in function 'forward'
	./densecap/DenseCapModel.lua:281: in function 'forward_test'
	run_model.lua:62: in function 'run_image'
	run_model.lua:149: in main chunk
	[C]: in function 'dofile'
	...ddra/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:150: in main chunk
	[C]: at 0x55f24f1f4570

WARNING: If you see a stack trace below, it doesn't point to the place where this error occurred. Please use only the one above.
stack traceback:
	[C]: in function 'error'
	/home/ruddra/torch/install/share/lua/5.1/nn/Container.lua:67: in function 'rethrowErrors'
	/home/ruddra/torch/install/share/lua/5.1/nn/Sequential.lua:44: in function 'forward'
	./densecap/DenseCapModel.lua:237: in function 'forward'
	./densecap/DenseCapModel.lua:281: in function 'forward_test'
	run_model.lua:62: in function 'run_image'
	run_model.lua:149: in main chunk
	[C]: in function 'dofile'
	...ddra/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:150: in main chunk
	[C]: at 0x55f24f1f4570

Attached image that we are running with

"Joint Common Sense and Relation Reasoning for Dense Relational Captioning"

This article may be suspected of plagiarism. If confirmed, you can write an email to IEEE to report plagiarism.

Dear scholar,
I found I can run the densecap model with -gpu -1 and it didn't throw error.

But your model set -gpu -1 didn't work.?

Hello, Thank you for your paper and code!

Currently I'm running your model and experimenting.

In many generative models, it is possible to manipulate the maximum length of outputs or adjust the number of outputs, where can this be done in run_model.lua?

Hi, it is an excellent research.
How can I run it win10 ?
How can I run it python3 ?
Do you have python3 version?

I am an amateur who likes DenseCap and I am interested in your novel ideas. However ,I only know something about python. So , do you have another python version?thank you

Nothing

Thanks for your great work!

I want to know if there is a script to produce the captions and region like Figure 4 in the paper?

hello! thanks for your code~~Could you tell me how long you've been training the whole network?And how many GPU you use to run it? I think that the network is too large that maybe I cannot run it. I haven't run it yet.Looking forward to your reply!thanks!

Hello! I do not find this file 'data/VG-regions-dicts_R2longv3.json' after running 'download.sh'. Could you tell me how can i solve this ?

some luarocks software package can be installed now，the code cann't be repetition

hello,when i run evaluate_model.lua,i meet a error: THCudaCheck FAIL file=........error=8:invalid device function..., and after searching it I can't solve it. my GPU is nvidia Telsa K80 ,while yours is TITAN X ,so ,what should i do to fix it?
thx. Looking forward to your reply~~

Hello to the maintainers of this repository. I have been trying to use this code but it is such a big pain to install torch and lua on my Ubuntu 22.04. I strongly suggest you guys to use pytorch instead to write your code. It is way more convenient and accessible. Please help people like me out and release the pytorch code. I will have to see if there are any other models available on the internet for image captioning. What is even the point of having such a good model if no-one can use it.

Hi, very interesting work.
How can I test some raw images out of the training dataset?

Could make the code work in the current mainstream environment in academia, based on Anaconda Python, Pytorch?