yufengm / adaptive Goto Github PK

is this project completely implement the result of the paper?

import argparse
from PIL import Image
import torch
import matplotlib.pyplot as plt
import numpy as np
import cv2
import pickle
from utils import CocoEvalLoader, to_var, show_images
from adaptive import Encoder2Decoder
from build_vocab import Vocabulary
from torch.autograd import Variable
from torchvision import transforms
def main():
pretrained = 'models/adaptive-1.pkl'
vocab_path = './data/vocab.pkl'
with open(vocab_path, 'rb') as f:
vocab = pickle.load(f)
# Define model and load pretrained
model = Encoder2Decoder(256, len(vocab), 512)
model.load_state_dict(torch.load(pretrained, map_location={'cuda:1':'cuda:0'}))
model.eval()
# Image transformation
transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.485, 0.456, 0.406),
(0.229, 0.224, 0.225))])
image = Image.open('./data/test/1.jpg')
image = image.resize([224, 224], Image.LANCZOS)
image = transform(image).unsqueeze(0)
image_tensor = Variable(image, volatile=True)
generated_captions, _, _ = model.sampler(image_tensor)
captions = generated_captions.data.numpy()
sampled_ids = captions[0]
sampled_caption = []
for word_id in sampled_ids:
word = vocab.idx2word[word_id]
if word == '':
break
else:
sampled_caption.append(word)
sentence = ' '.join(sampled_caption)
# return sentence
print (sentence)

if name == 'main':
main()
出错：TypeError: torch.index_select received an invalid combination of arguments - got (torch.FloatTensor, int, !torch.cuda.LongTensor!)

please help! Thanks

Hi, first of all thank you for the great repo! I noticed that you are using different tokenizers for the training data - nltk.tokenize.word_tokenize in build_vocab.py - and the validation data - PTBTokenizer from coco/pycocoevalcap/tokenizer/ptbtokenizer.py. The first one doesn't split on punctuation, while the second one does, which leads to many <unk> tokens for captions that have hyphenated words. It would be great if you could have both training and captioning use the same tokenizer. Thanks!

please help me how to find accuracy for each epochs ? Is their any code ?

Its showing Cross Entropy Loss value and Perplexity value in the training !!

Hi, there is a 'batch_size_t'，which will filter the samples that have short length, but why do this, I dont think this will work, maybe this will lead the model adopt very few samples?
batch_size_t = sum([l > timestep for l in decode_lengths]) current_input = inputs[:batch_size_t, timestep, :]

Please help me, thanks.

Thanks for your work on this pytorch version of 'knowing when to look'. Can you public the pre-trained model?

Could you please upload a example script?

could you tell me how to test ?

I got CIDer 0.82 only ，could you please help me about how I can imporve the score?thanks

Thanks for your contribution!
But I have a question that during training, the total number of iteration of every epoch is 9446. I am wondering how this comes from...As we know, there are about 80k+40k-10k=110k images in the training set.
So the length of data_loader should be : 110k/batch_size=110k/60=1833, rather than 9446.
Can anyone help me?

Traceback (most recent call last):
File "train.py", line 9, in
from utils import coco_eval, to_var
File "/data/DATA_DIR/utils.py", line 12, in
from coco.pycocoevalcap.eval import COCOEvalCap
File "/data/DATA_DIR/coco/pycocoevalcap/eval.py", line 3, in
from .bleu.bleu import Bleu
File "/data/DATA_DIR/coco/pycocoevalcap/bleu/bleu.py", line 11, in
from bleu_scorer import BleuScorer
ModuleNotFoundError: No module named 'bleu_scorer'

@yufengm
After one epoch, I want to calculate the METEOR score, but error as follow:
computing METEOR score...
Traceback (most recent call last):
File "train_model.py", line 268, in
main(args)
File "train_model.py", line 177, in main
method_score = coco_eval(adaptive, args, epoch)
File "/home1/haoyanlong/imagecaption/Adaptive_Attention/code/utils.py", line 173, in coco_eval
cocoEval.evaluate()
File "/home1/haoyanlong/imagecaption/Adaptive_Attention/code/coco/pycocoevalcap/eval.py", line 51, in evaluate
score, scores = scorer.compute_score(gts, res)
File "/home1/haoyanlong/imagecaption/Adaptive_Attention/code/coco/pycocoevalcap/meteor/meteor.py", line 38, in compute_score
stat = self._stat(res[i][0], gts[i])
File "/home1/haoyanlong/imagecaption/Adaptive_Attention/code/coco/pycocoevalcap/meteor/meteor.py", line 56, in _stat
self.meteor_p.stdin.write('{}\n'.format(score_line))
IOError: [Errno 32] Broken pipe

Hello,

When I try to run the 'train.py' there comes this error.

Traceback (most recent call last):
File "train.py", line 263, in
main( args )
File "train.py", line 176, in main
cider = coco_eval( adaptive, args, epoch )
File "/home/harryjhnam/Documents/final_project/Adaptive/utils.py", line 120, in coco_eval
CocoEvalLoader( args.image_dir, args.caption_val_path, transform ).samples
AttributeError: 'CocoEvalLoader' object has no attribute 'samples'

I have modified your code for python3.5.2 and also tried on the python2.7 with your original code, but both methods occur the same error.

Hi,

I trained a fine-tuned model on my own custom dataset, and the results are generally quite accurate. However, with some images I've noticed that if I sample the same image multiple times I get different captions. The differences are usually small, but for my use case I need consistency. I tried to debug by printing out the encoder tensors and it looks like the model encodes the same image differently at different samplings. Is this expected behavior? Is there a way to "stabilize" the encoder so it encodes the same image the same way each time?

Thanks!

Hi, this is the best project, but when i use the train.py the error occurs.
Here is the error
RuntimeError: matrix and matrix expected at /opt/conda/conda-bld/pytorch_1501971235237/work/pytorch-0.1.12/torch/lib/THC/generic/THCTensorMathBlas.cu:237
My pytorch version is 0.1.12
Best,

I'm pretty sure I'm using a free GPU.
Thanks in advance.

Could please tell me how to solve this runtime error ?

Tensor flow dimension

Hi,

I am training your model with MSCOCO dataset, and validating with Flickr validation data.

After 2nd epoch, I started to have results like;

"A yellow train", "A small bus", "An aeroplane." etc.

The remaining of the sentences are missing.

Is it because of the data, or what?

Thank you,