Python misc.utils.decode_sequence() Examples

The following are 2 code examples of misc.utils.decode_sequence(). You can vote up the ones you like or vote down the ones you don't like, and go to the original project or source file by following the links above each example. You may also want to check out all available functions/classes of the module misc.utils , or try the search function .
Example #1
Source File: eval.py    From video-caption.pytorch with MIT License 5 votes vote down vote up 
def test(model, crit, dataset, vocab, opt):
    model.eval()
    loader = DataLoader(dataset, batch_size=opt["batch_size"], shuffle=True)
    scorer = COCOScorer()
    gt_dataframe = json_normalize(
        json.load(open(opt["input_json"]))['sentences'])
    gts = convert_data_to_coco_scorer_format(gt_dataframe)
    results = []
    samples = {}
    for data in loader:
        # forward the model to get loss
        fc_feats = data['fc_feats'].cuda()
        labels = data['labels'].cuda()
        masks = data['masks'].cuda()
        video_ids = data['video_ids']
      
        # forward the model to also get generated samples for each image
        with torch.no_grad():
            seq_probs, seq_preds = model(
                fc_feats, mode='inference', opt=opt)

        sents = utils.decode_sequence(vocab, seq_preds)

        for k, sent in enumerate(sents):
            video_id = video_ids[k]
            samples[video_id] = [{'image_id': video_id, 'caption': sent}]

    with suppress_stdout_stderr():
        valid_score = scorer.score(gts, samples, samples.keys())
    results.append(valid_score)
    print(valid_score)

    if not os.path.exists(opt["results_path"]):
        os.makedirs(opt["results_path"])

    with open(os.path.join(opt["results_path"], "scores.txt"), 'a') as scores_table:
        scores_table.write(json.dumps(results[0]) + "\n")
    with open(os.path.join(opt["results_path"],
                           opt["model"].split("/")[-1].split('.')[0] + ".json"), 'w') as prediction_results:
        json.dump({"predictions": samples, "scores": valid_score},
                  prediction_results)
Example #2
Source File: eval.py    From video-caption.pytorch with MIT License 4 votes vote down vote up 
def test(model, crit, dataset, vocab, opt):
    model.eval()
    loader = DataLoader(dataset, batch_size=opt["batch_size"], shuffle=False)
    scorer = COCOScorer()
    gt_dataframe = json_normalize(
        json.load(open(opt["input_json"]))['sentences'])
    gts = convert_data_to_coco_scorer_format(gt_dataframe)
    #results = []
    samples = {}
    for index, data in enumerate(loader):
        print 'batch: '+str((index+1)*opt["batch_size"])
        # forward the model to get loss
        fc_feats = Variable(data['fc_feats'], volatile=True).cuda()
        labels = Variable(data['labels'], volatile=True).long().cuda()
        masks = Variable(data['masks'], volatile=True).cuda()
        video_ids = data['video_ids']
      
        # forward the model to also get generated samples for each image
        seq_probs, seq_preds = model(
            fc_feats, mode='inference', opt=opt)
        # print(seq_preds)

        sents = utils.decode_sequence(vocab, seq_preds)

        for k, sent in enumerate(sents):
            video_id = video_ids[k]
            samples[video_id] = [{'image_id': video_id, 'caption': sent}]
        # break
    with suppress_stdout_stderr():
        valid_score = scorer.score(gts, samples, samples.keys())
    #results.append(valid_score)
    #print(valid_score)

    if not os.path.exists(opt["results_path"]):
        os.makedirs(opt["results_path"])
    result = OrderedDict()
    result['checkpoint'] = opt["saved_model"][opt["saved_model"].rfind('/')+1:]
    score_sum = 0
    for key, value in valid_score.items():
        score_sum += float(value)
    result['sum'] = str(score_sum)
    #result = OrderedDict(result, **valid_score)
    result = OrderedDict(result.items() + valid_score.items())
    print result
    if not os.path.exists(opt["results_path"]):
        os.makedirs(opt["results_path"])
    with open(os.path.join(opt["results_path"], "scores.txt"), 'a') as scores_table:
        scores_table.write(json.dumps(result) + "\n")
    with open(os.path.join(opt["results_path"],
                           opt["model"].split("/")[-1].split('.')[0] + ".json"), 'w') as prediction_results:
        json.dump({"predictions": samples, "scores": valid_score},
                  prediction_results)