Awesome Open Source
Awesome Open Source

Tensorflow Implementation of Deeper LSTM+ normalized CNN for Visual Question Answering

Provide tensorflow edition for VQA_LSTM_CNN, training a deeper LSTM and normalized CNN Visual Question Answering model. The current code can reach same accuracy with original torch code on Open-Ended (use COCO train set to train and validation set to evaluation). You can check original torch edtion from VQA_LSTM_CNN for more details.

Requirements

This code is written in Python and requires Tensorflow. The preprocssinng code is in Python.

Prepare Data (from VQA_LSTM_CNN)

(Here's a copy from the original readme.md) The first thing you need to do is to download the data and do some preprocessing. Head over to the data/ folder and run

$ python vqa_preprocessing.py --download True --split 1

--download Ture means you choose to download the VQA data from the VQA website and --split 1 means you use COCO train set to train and validation set to evaluation. --split 2 means you use COCO train+val set to train and test set to evaluate. After this step, it will generate two files under the data folder. vqa_raw_train.json and vqa_raw_test.json

Once you have these, we are ready to get the question and image features. Back to the main folder, run

$ python prepro.py --input_train_json data/vqa_raw_train.json --input_test_json data/vqa_raw_test.json --num_ans 1000

to get the question features. --num_ans specifiy how many top answers you want to use during training. You will also see some question and answer statistics in the terminal output. This will generate two files in your main folder, data_prepro.h5 and data_prepro.json. To get the image features, run

$ th prepro_img.lua -input_json data_prepro.json -image_root path_to_image_root -cnn_proto path_to_cnn_prototxt -cnn_model path to cnn_model

Here we use VGG_ILSVRC_19_layers model. After this step, you can get the image feature data_img.h5. We have prepared everything and ready to launch training. You can simply run

Training and Testing

To train on the prepared dataset, comment out test(). We simply run the program with python.

$ python model_VQA.py

with the default parameter, this will take several hours and will generate the model under model_save To test, comment out train() and run the same program, this will generate data.json We finally run a simple program to correct the generated json files.

$ python s2i.py

This will generate the result OpenEnded_mscoco_lstm_results.json. To evaluate the accuracy of generate result, you need to download the VQA evaluation tools.


Get A Weekly Email With Trending Projects For These Topics
No Spam. Unsubscribe easily at any time.
python (52,053
tensorflow (2,131
question-answering (104
vqa (23

Find Open Source By Browsing 7,000 Topics Across 59 Categories