priyamtejaswin / devise-keras

Interpretable Image Search by Priyam Tejaswin and Akshay Chawla
http://35.227.39.159:5050/
22 stars 9 forks source link

Master Branch, followed SETUP steps, get Model Dimension Mis-match error #35

Open jkent42 opened 5 years ago

jkent42 commented 5 years ago

SETUP

edit local.cfg to set LOCAL/PROD in ENV

bash SETUP.sh

python scrape_and_save_images.py

python extract_word_embeddings.py

python shuffle_val_data.py

Ensure DS_Store files are not in the image folders.

python clean_data.py TRAIN

python clean_data.py VALID

python extract_features_and_dump.py TRAIN

// GET A MIS-ALIGNMENT ERROR// when running "python extract_features_and_dump.py TRAIN ..........

File "/usr/lib/python2.7/site-packages/tensorflow/python/framework/common_shapes.py", line 610, in call_cpp_shape_fn debug_python_shape_fn, require_shape_fn) File "/usr/lib/python2.7/site-packages/tensorflow/python/framework/common_shapes.py", line 676, in _call_cpp_shape_fn_impl raise ValueError(err.message) ValueError: Dimension 0 in both shapes must be equal, but are 3 and 64 for 'Assign' (op: 'Assign') with input shapes: [3,3,224,64], [64,3,3,3].

prior to this error, I got another error:

ValueError: Negative dimension size caused by subtracting 2 from 1 for 'block2_pool/MaxPool' (op: 'MaxPool') with input shapes: [?,1,112,128].

to attempt to address this error, I added parameter dim_ordering="th" in the MaxPooling2D calls.

Please advise what would be good steps to address these errors.

Is there an update for model building in python_extract_features_and_dump.py ?

jkent42 commented 5 years ago

Not sure this is the correct branch (looking at the master branch) if the goal is building the basic DeViSE model. Other branches include command line parameters (eg file names) for the python programs called in the setup steps. Unless file names are hard coded. I will look further. Alternatively, I may abandon my attempt to work with this code. Developers did not respond to email. Responded to issue raised about demo server being down - sounds like it will not be coming back up. That makes this essentially an abandoned project. That is unfortunate, as a lot of solid work has gone into this effort.