Final commit

Author: surajsoni5

README.md

@@ -109,7 +109,7 @@ You can install these dependencies using `pip install -r requirements.txt`
 ```
 python datapreprocess.py \path\to\data\set \path\to\vocab\
 python train.py
-python train_justify.py
+python train_justify.py 
 ```
 #### Testing
 Download the pretrained models [checkpoint_d](https://drive.google.com/open?id=1w4zF82hgbPmU9hAHsY92myjOU6HYD_BI) and [checkpoint_j](https://drive.google.com/open?id=1QLyqU5HZHYAyRJTSsIBPqG8zARU0oZUL)

beamsearch.py

@@ -417,7 +417,7 @@ def beam_search_discriminative(encoder, decoder, image_path_t,image_path_d,lambd
     elif sys.argv[1]=="cj": ## cj Image_path target_class distractor class
         checkpoints=torch.load('checkpoint_j')
         encoder=checkpoints['encoder']
-        decoder=checkpoints['decoder']
+        decoder=checkpoints['decoder'] 
         encoder.eval()
         decoder.eval()
         image_path=sys.argv[2]

datapreprocess.py

@@ -7,7 +7,7 @@
 import torchfile as tf
 from PIL import Image
 from matplotlib.pyplot import imshow
-import sys
+import sys    
 cub_dataset    = sys.argv[1] #'C:/Users/hello/Desktop/Accads/CUB_200_2011/' # Location of cub dataset 
 captions_dataset= sys.argv[2] # 'C:/Users/hello/Desktop/Accads/cvpr2016_cub/word_c10/' # Location of captions dataset
 VocabSize=5725+1 # Starts from 1 and ends at 5725, include start token also

datasets.py

@@ -5,7 +5,7 @@
 from torch.utils.data import Dataset
 import h5py
 import numpy as np
-
+  
 class CaptionDataset(Dataset):
     def __init__(self, transform):
         self.h = h5py.File('dataset.hdf5','r',driver='core')
@@ -14,7 +14,7 @@ def __init__(self, transform):
         self.captions= np.array(self.h['captions'])
         self.class_k =np.array(self.h['class'])
         self.maxcapslen=30+3 
-        self.numcaptions=self.h['numcaptions'].value
+        self.numcaptions=self.h['numcaptions'].value 
         self.transform=transform
 
     def __getitem__(self, i):

models.py

@@ -21,7 +21,7 @@ def forward(self, images):
         out = out.permute(0,2,3,1) # Batch* d*e*e to batch*e*e*d, helps in attention
         return out
 
-class Attention(nn.Module): 
+class Attention(nn.Module):  
     def __init__(self, encoder_dim, decoder_dim, attention_dim):
         super(Attention, self).__init__()
         self.encoder_att = nn.Linear(encoder_dim, attention_dim)  

train.py

@@ -14,7 +14,7 @@
 dropout = 0.5 # Dropout rate
 decoder_lr = 2*1e-3 # Decoder learning rate
 numepochs=100 # Number of epochs
-load=False ## Make this false when you don't want load a checkpoint
+load=False ## Make this false when you don't want load a checkpoint 
 
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 

train_justify.py

@@ -14,7 +14,7 @@
 dropout = 0.5 # Dropout rate
 decoder_lr = 2*1e-3 # Decoder learning rate
 numepochs=100 # Number of epochs
-load=False ## Make this false when you don't want load a checkpoint
+load=False ## Make this false when you don't want load a checkpoint 
 
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")