hlamba28/automatic-image-captioning

Generating Captions for images using Deep Learning

automatic-image-captioning's Issues

Issue in loading train/test images in a list

I am not able to load train/test images in a list. It shows empty list.

changing learning rate after 20 epochs error

code uses a direct method to change the learning rate of the model to 0.0001 which raises an error for reason unknown. A quick fix would be to use keras backend functionality to change the learning rate.
Instead of model.optimizer.lr = 0.0001 use K.set_value(model.optimizer.lr, 0.0001) after importing keras.backend as K

Train model

It takes so much time to train the model. Can someone please provide me 'model.h5' file(pre-trained model)

randomization/shuffling in data generator

Hi, I read your blog post, which I found really helpful for learning about the image captioning problem. But I was curious about the lack of randomization or shuffling of the data in the generator. Do you see this as a potential shortcoming, or is there a reason that it isn't necessary in this problem?

Missing test_img and train_img

these lists are not defined in your code, i do not think this code works as of yet

import error

pickle is not imported. It causes error while trying to use pickle.dump function. A simple import pickle would suffice.

data_generator error prevents training

There is an error in the data_generator function that, when trying to train the model, it will throw the following error:

could not broadcast input array from shape (168,2048) into shape (168)

The project was run from the cloned repo, dataset was copied and all paths are checked. All cell outputs until the training cell are equal to the original notebook.

Undefined Variable

Hello,
I went ahead and tried to make your tutorial work. It has a variable called z that idk what it does and it's not defined before the line it's used, tried to assign integers but I get errors during greedy search (naturally).
Can you help me out?

z+=1
pic = list(encoding_test.keys())[z]
image = encoding_test[pic].reshape((1,2048))
x=plt.imread(images+pic)
plt.imshow(x)
plt.show()
print("Greedy:",greedySearch(image))

Loss value for the model

Can you please tell me what was the best loss value (on validation set) you were getting and after which epoch?

Thanks in advance.

Less accuracy

I have used same code and same dataset but getting very much less accuracy. Can you please help with issue.