ranjaykrishna / iq Goto Github PK
View Code? Open in Web Editor NEWInformation Maximizing Visual Question Generation
Home Page: https://cs.stanford.edu/people/ranjaykrishna/iq/index.html
License: Other
Information Maximizing Visual Question Generation
Home Page: https://cs.stanford.edu/people/ranjaykrishna/iq/index.html
License: Other
Hi! When running train_iq.py, I couldn't find vae_val_dataset_weights.json. Do you know where I can download or generate this file?
In line 186 of train_iq.py, perhaps "scheduler" should be changed to "info_scheduler"? Otherwise, it seems that "info_scheduler" has not been updated continuously? This may be the reason why the loss gradually becomes NaN during the training process.
The inference script provided in the repo doesn't seem to work in its current form. I had to make some changes to it to make it run. In doing so, I am not able to replicate the results provided in the paper. The quantitative metrics give a much lower score in comparison to what is reported in the paper.
Please help us point out the mistake we are doing or update the inference/training file with the correct one.
Thank you.
Hi @ranjaykrishna ,
i am trying to execute python utils/vocab.py getting an error
"Traceback (most recent call last):
File "utils/vocab.py", line 144, in
vocab = build_vocab(args.questions, args.cat2ans, args.threshold)
File "utils/vocab.py", line 93, in build_vocab
answer = tokenize(answer.encode('utf8'))
AttributeError: 'dict' object has no attribute 'encode'"
can you help me to resolve this issue...
Thanks & Regards,
Manikantha Sekhar...
Hi Ranjay,
Would you please kindly fix this issue? Thanks a lot
Hello Ranjay, thanks for your work.
I have run the train_iq.py using the default lr 0.001, but after 4 epochs, the loss and other metrics got Nan. Is there something wrong with the training script?
Looking forward to your reply. Thanks again.
Is it possible to release the pretrained file?
Hi, Ranjay
Thanks for releasing the code.
The input parameters of "iq/utils/store_dataset.py" includes a path to the processed "ans2cat.json file". There isn't a processed code for generating this file. Can it be done by exchanging the position of key-value pair in here?
Thanks
Becky
I tried to train the model as written in readme.
However, when I tried to specify model_type, I got an error:
$ python train_iq.py --model-type ia2q
usage: train_iq.py [-h] [--model-path MODEL_PATH] [--crop-size CROP_SIZE] [--log-step LOG_STEP] [--save-step SAVE_STEP] [--eval-steps EVAL_STEPS]
[--eval-every-n-steps EVAL_EVERY_N_STEPS] [--num-epochs NUM_EPOCHS] [--batch-size BATCH_SIZE] [--num-workers NUM_WORKERS]
[--learning-rate LEARNING_RATE] [--info-learning-rate INFO_LEARNING_RATE] [--patience PATIENCE] [--max-examples MAX_EXAMPLES] [--lambda-gen LAMBDA_GEN]
[--lambda-z LAMBDA_Z] [--lambda-t LAMBDA_T] [--lambda-a LAMBDA_A] [--lambda-i LAMBDA_I] [--lambda-z-t LAMBDA_Z_T] [--vocab-path VOCAB_PATH]
[--dataset DATASET] [--val-dataset VAL_DATASET] [--train-dataset-weights TRAIN_DATASET_WEIGHTS] [--val-dataset-weights VAL_DATASET_WEIGHTS]
[--cat2name CAT2NAME] [--load-model LOAD_MODEL] [--rnn-cell RNN_CELL] [--hidden-size HIDDEN_SIZE] [--num-layers NUM_LAYERS] [--max-length MAX_LENGTH]
[--encoder-max-len ENCODER_MAX_LEN] [--bidirectional] [--use-glove] [--embedding-name EMBEDDING_NAME] [--num-categories NUM_CATEGORIES]
[--dropout-p DROPOUT_P] [--input-dropout-p INPUT_DROPOUT_P] [--num-att-layers NUM_ATT_LAYERS] [--z-size Z_SIZE] [--no-image-recon] [--no-answer-recon]
[--no-category-space]
train_iq.py: error: unrecognized arguments: --model-type ia2q
How can I successfully specify model_type
?
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. ๐๐๐
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.