"""Structure generator components in PCG"""
def __init__(self,
encoder=None,
decoder=None,
outViewN=8,
outW=128,
outH=128,
renderDepth=1.0):
super(Structure_Generator, self).__init__()
if encoder: self.encoder = encoder
else: self.encoder = Encoder()
if decoder: self.decoder = decoder
else: self.decoder = Decoder(outViewN, outW, outH, renderDepth)
def forward(self, x):
latent = self.encoder(x)
XYZ, maskLogit = self.decoder(latent)
return XYZ, maskLogit
# TESTING
if __name__ == '__main__':
import options
cfg = options.get_arguments()
encoder = Encoder()
decoder = Decoder(cfg.outViewN, cfg.outW, cfg.outH, cfg.renderDepth)
model = Structure_Generator()
import options
import utils
from trainer import TrainerStage2
if __name__ == "__main__":
print("=======================================================")
print("Train structure generator with joint 2D optimization from novel viewpoints")
print("=======================================================")
cfg = options.get_arguments()
EXPERIMENT = f"{cfg.model}_{cfg.experiment}"
MODEL_PATH = f"models/{EXPERIMENT}"
LOG_PATH = f"logs/{EXPERIMENT}"
utils.make_folder(MODEL_PATH)
utils.make_folder(LOG_PATH)
criterions = utils.define_losses()
dataloaders = utils.make_data_novel(cfg)
model = utils.build_structure_generator(cfg).to(cfg.device)
optimizer = utils.make_optimizer(cfg, model)
scheduler = utils.make_lr_scheduler(cfg, optimizer)
logger = utils.make_logger(LOG_PATH)
writer = utils.make_summary_writer(EXPERIMENT)
def on_after_epoch(model, df_hist, images, epoch, saveEpoch):
utils.save_best_model(MODEL_PATH, model, df_hist)
.astype(np.int)
# Convert to Tensor
images = torch.from_numpy(images).permute((0, 3, 1, 2))
depthGT = torch.from_numpy(depthGT).permute((0, 3, 1, 2))
maskGT = torch.from_numpy(maskGT).permute((0, 3, 1, 2))
return {
"inputImage": images,
"depthGT": depthGT,
"maskGT": maskGT,
}
if __name__ == "__main__":
import options
CFG = options.get_arguments()
ds_fixed = PointCloud2dDataset(CFG)
dl_fixed = DataLoader(ds_fixed,
batch_size=CFG.chunkSize,
shuffle=False,
collate_fn=ds_fixed.collate_fn_fixed)
ds_novel = PointCloud2dDataset(CFG, loadNovel=True)
dl_novel = DataLoader(ds_novel,
batch_size=CFG.chunkSize,
shuffle=False,
collate_fn=ds_novel.collate_fn)
images = batch_n["image_in"][modelIdx, angleIdx]
depthGT = np.transpose(batch_n["depth"][modelIdx], axes=[0, 2, 3, 1])
maskGT = np.transpose(batch_n["mask"][modelIdx], axes=[0, 2, 3, 1])\
.astype(np.int)
# Convert to Tensor
images = torch.from_numpy(images).permute((0,3,1,2))
depthGT = torch.from_numpy(depthGT).permute((0,3,1,2))
maskGT = torch.from_numpy(maskGT).permute((0,3,1,2))
return {
"inputImage": images,
"depthGT": depthGT,
"maskGT": maskGT,
}
if __name__ == "__main__":
import options
CFG = options.get_arguments()
ds_fixed = PointCloud2dDataset(CFG)
dl_fixed = DataLoader(ds_fixed, batch_size=CFG.chunkSize,
shuffle=False, collate_fn=ds_fixed.collate_fn_fixed)
ds_novel = PointCloud2dDataset(CFG, loadNovel=True)
dl_novel = DataLoader(ds_novel, batch_size=CFG.chunkSize,
shuffle=False, collate_fn=ds_novel.collate_fn)
# d = compute_distance(sketch_features.copy(), shape_features.copy(), l2=False)
# scio.savemat('test/example.mat',{'d':d, 'feat':dataset_features, 'labels':dataset_labels})
# AUC, mAP = map_and_auc(sketch_labels.copy(), shape_labels.copy(), d)
# print(' * Feature AUC {0:.5} mAP {0:.5}'.format(AUC, mAP))
d_feat = compute_distance(sketch_features.copy(), shape_features.copy(), l2=False)
d_feat_norm = compute_distance(sketch_features.copy(), shape_features.copy(), l2=True)
mAP_feat = compute_map(sketch_labels.copy(), shape_labels.copy(), d_feat)
mAP_feat_norm = compute_map(sketch_labels.copy(), shape_labels.copy(), d_feat_norm)
print(' * Feature mAP {0:.5%}\tNorm Feature mAP {1:.5%}'.format(mAP_feat, mAP_feat_norm))
d_score = compute_distance(sketch_scores.copy(), shape_scores.copy(), l2=False)
mAP_score = compute_map(sketch_labels.copy(), shape_labels.copy(), d_score)
d_score_norm = compute_distance(sketch_scores.copy(), shape_scores.copy(), l2=True)
mAP_score_norm = compute_map(sketch_labels.copy(), shape_labels.copy(), d_score_norm)
if opt.sf:
shape_paths = [img[0] for img in shape_dataloader.dataset.shape_target_path_list]
sketch_paths = [img[0] for img in sketch_dataloader.dataset.sketch_target_path_list]
scio.savemat('{}/test_feat_temp.mat'.format(opt.checkpoint_folder), {'score_dist':d_score, 'score_dist_norm': d_score_norm, 'feat_dist': d_feat, 'feat_dist_norm': d_feat_norm,'sketch_features':sketch_features, 'sketch_labels':sketch_labels, 'sketch_scores': sketch_scores,
'shape_features':shape_features, 'shape_labels':shape_labels, 'sketch_paths':sketch_paths, 'shape_paths':shape_paths})
print(' * Score mAP {0:.5%}\tNorm Score mAP {1:.5%}'.format(mAP_score, mAP_score_norm))
return [sketch_top1.avg, shape_top1.avg, mAP_feat, mAP_feat_norm, mAP_score, mAP_score_norm]
if __name__ == '__main__':
from options import get_arguments
opt = get_arguments()
main(opt)
def main():
args = get_arguments()
random.seed(args.random_seed)
np.random.seed(args.random_seed) # sklearn use np to generate random value
# Create folders and set logging format
args.model_dir = os.path.join(args.out_dir, 'ckpt-{}'.format(args.class_weight_scheme))
args.log_dir = os.path.join(args.out_dir, 'log')
args.ensemble_dir = os.path.join(args.out_dir, 'ensemble-{}'.format(args.class_weight_scheme))
if args.class_weight_scheme == 'customize':
args.model_dir = os.path.join(args.model_dir, 'weight{}'.format(args.additional_weight))
args.ensemble_dir = os.path.join(args.ensemble_dir, 'weight{}'.format(args.additional_weight))
prepare_folders(args)
logger = set_logging(args)
logger.info("Here is the arguments of this running:")
logger.info("{}".format(args))
utils.check_args_conflict(args)
# Set files which contain data for training and test. If use "trecis2019-A", it means we want to tune parameters.
args.data_prefix = "trecis2019-B"
# Note that for 2019-B submission, all '2019' means '2019-B' and '2018' means '2018 + 2019-A'
label_file = os.path.join(args.data_dir, 'ITR-H.types.v{}.json'.format(
4 if args.data_prefix == "trecis2019-B" else 3))
tweet_file_list = [os.path.join(args.data_dir, 'all-tweets.txt')]
tweet_file_list_2019 = [os.path.join(args.data_dir, 'all-tweets-2019.txt')]
train_file_list = [os.path.join(args.data_dir, 'TRECIS-CTIT-H-Training.json')]
train_file_list += [os.path.join(args.data_dir, 'TRECIS-2018-TestEvents-Labels',
'assr{}.test'.format(i)) for i in range(1, 7)]
if args.data_prefix == "trecis2019-B":
train_file_list += [os.path.join(args.data_dir, '2019ALabels', '2019A-assr{}.json'.format(i)) for i in range(1, 6)]
train_file_list += [os.path.join(args.data_dir, '2019ALabels', '2019-assr2.json')]
test_raw_tweets_json_folder = 'download_tweets'
# Some output files which has been formalized for further usages.
formal_train_file = os.path.join(args.data_dir, 'train.txt{}'.format('_small' if args.sanity_check else ''))
formal_test_file = os.path.join(args.data_dir, 'test.txt{}')
tweet_text_out_file = os.path.join(args.out_dir, 'tweets-clean-text.txt')
tweet_id_out_file = os.path.join(args.out_dir, 'tweets-id.txt')
tweet_text_out_file_2019 = os.path.join(args.out_dir, 'tweets-clean-text-2019.txt')
tweet_id_out_file_2019 = os.path.join(args.out_dir, 'tweets-id-2019.txt')
predict_priority_score_out_file = os.path.join(args.out_dir, 'predict_priority_score.txt')
# Set files for submission.
args.model_name = '{0}{1}'.format(args.model, '-event' if args.event_wise else '')
args.dev_label_file = os.path.join(args.ensemble_dir, 'dev_label.txt')
args.dev_predict_file = os.path.join(args.ensemble_dir, 'dev_predict_{}.txt'.format(args.model_name))
args.test_predict_file = os.path.join(args.ensemble_dir, 'test_predict_{}.txt'.format(args.model_name))
args.submission_folder = utils.prepare_submission_folder(args)
args.submission_file = os.path.join(args.submission_folder, 'submission_{}'.format(args.model_name))
# As the original files provided by TREC is quite messy, we formalize them into train and test file.
utils.formalize_files(train_file_list, formal_train_file, args)
utils.formalize_test_file(test_raw_tweets_json_folder, formal_test_file, prefix=args.data_prefix)
logger.info("The training data file is {0} and testing data file is {1}".format(
formal_train_file, formal_test_file))
# Step0. Extract some info which can be used later (also useful for generating submission files).
label2id, majority_label, short2long_label = utils.get_label2id(label_file, formal_train_file, args.cv_num)
id2label = utils.get_id2label(label2id)
class_weight = utils.get_class_weight(args, label2id, id2label, formal_train_file)
# When get submission, there is no need to run all following steps, but only read the `test_predict_file` and
# pick some classes as final output according to policy (such as top-2 or auto-threshold).
# You MUST run `--predict_mode` in advance to get the `test_predict_file` prepared.
if args.get_submission:
postpro = PostProcess(args, label2id, id2label, class_weight, majority_label, short2long_label,
formal_train_file, formal_test_file, test_raw_tweets_json_folder,
predict_priority_score_out_file)
postpro.pick_labels_and_write_final_result()
quit()
# Step1. Preprocess and extract features for all tweets
tweetid_list, tweet_content_list = utils.get_tweetid_content(tweet_file_list)
utils.write_tweet_and_ids(tweetid_list, tweet_content_list, tweet_text_out_file, tweet_id_out_file)
tweetid_list_2019, tweet_content_list_2019 = utils.get_tweetid_content(tweet_file_list_2019)
utils.write_tweet_and_ids(tweetid_list_2019, tweet_content_list_2019, tweet_text_out_file_2019,
tweet_id_out_file_2019)
# Note that before `extract_features()`, we should manually run the `extract_features.sh` in `feature_tools`.
# quit() # The `extract_features.sh` only need to be run once for the same dataset.
preprocess = Preprocess(args, tweetid_list, tweet_content_list, label2id, tweet_id_out_file)
preprocess.extract_features()
preprocess_2019 = Preprocess(args, tweetid_list_2019, tweet_content_list_2019, label2id,
tweet_id_out_file_2019, test=True)
preprocess_2019.extract_features()
if args.train_regression:
data_x, data_score = preprocess.extract_train_data(formal_train_file, get_score=True)
train_regression = TrainRegression(args, data_x, data_score)
if args.cross_validate:
train_regression.cross_validate()
quit()
if args.cross_validate:
# Step2. Train and Cross-validation (for tuning hyper-parameters).
# If we want to do ensemble in the future, we need the prediction on dev data by setting `--cross_validate`.
if args.event_wise:
data_x, data_y, event2idx_list, line_num = preprocess.extract_train_data(formal_train_file)
data_predict_collect = np.zeros([line_num, len(label2id)])
metrics_collect = []
metric_names = None
for event_type in utils.idx2event_type:
it_data_x, it_data_y = data_x[event_type], data_y[event_type]
train = Train(args, it_data_x, it_data_y, id2label, preprocess.feature_len, class_weight, event_type)
metrics, predict_score = train.train()
for i, idx in enumerate(event2idx_list[event_type]):
data_predict_collect[idx] = predict_score[i]
metrics_collect.append((metrics, it_data_x.shape[0]))
if metric_names is None:
metric_names = train.metric_names
utils.get_final_metrics(metrics_collect, metric_names)
else:
data_x, data_y = preprocess.extract_train_data(formal_train_file)
train = Train(args, data_x, data_y, id2label, preprocess.feature_len, class_weight)
_, data_predict_collect = train.train()
if args.predict_mode:
utils.write_predict_and_label(args, formal_train_file, label2id, data_predict_collect)
if args.predict_mode:
# Step3. Get the 2019 test data, and retrain the model on all training data, then predict on the 2019-test
if args.event_wise:
data_x, data_y, _, _ = preprocess.extract_train_data(formal_train_file)
test_x, event2idx_list, line_num = preprocess_2019.extract_formalized_test_data(formal_test_file)
test_predict_collect = np.zeros([line_num, len(label2id)])
for event_type in utils.idx2event_type:
it_data_x, it_data_y, it_test_x = data_x[event_type], data_y[event_type], test_x[event_type]
if len(it_test_x) == 0:
print("[WARNING] There are no event belongs to {} for the test data".format(event_type))
continue
train = Train(args, it_data_x, it_data_y, id2label,
preprocess_2019.feature_len, class_weight, event_type)
train.train_on_all()
predict_score = train.predict_on_test(it_test_x)
for i, idx in enumerate(event2idx_list[event_type]):
test_predict_collect[idx] = predict_score[i]
else:
data_x, data_y = preprocess.extract_train_data(formal_train_file)
test_x = preprocess_2019.extract_formalized_test_data(formal_test_file)
train = Train(args, data_x, data_y, id2label, preprocess_2019.feature_len, class_weight)
train.train_on_all()
test_predict_collect = train.predict_on_test(test_x)
utils.write_predict_res_to_file(args, test_predict_collect)
if args.train_regression:
test_x = preprocess_2019.extract_formalized_test_data(formal_test_file)
if args.event_wise:
# For event_wise setting, there will be many additional things extracted, what we need is only test_x.
test_x = test_x[0]
train_regression.train()
predict_priority_score = train_regression.predict_on_test(test_x)
utils.write_predict_score_to_file(predict_priority_score, predict_priority_score_out_file)
if args.ensemble is not None:
# TODO(junpeiz): Average the priority score for ensemble.
# Step4 (optional). Do the ensemble of different model
if args.event_wise:
raise NotImplementedError("We don't want to ensemble for event-wise models")
else:
out_file = os.path.join(args.out_dir, 'ensemble_out.txt')
# Note the file list contains predictions from all models with and without the '-event' suffix.
# So, we need to train both event-wise and not event-wise models or just delete those files in the folder.
dev_predict_file_list = utils.get_predict_file_list(args.ensemble_dir, 'dev_predict_')
test_predict_file_list = utils.get_predict_file_list(args.ensemble_dir, 'test_predict_')
train_x = utils.get_ensemble_feature(dev_predict_file_list)
train_y = utils.get_ensemble_label(args.dev_label_file)
print("The shape of ensemble train_x is {0}".format(train_x.shape))
utils.ensemble_cross_validate(train_x, train_y, id2label, train.mlb, args.ensemble)
test_x = utils.get_ensemble_feature(test_predict_file_list)
predict = utils.ensemble_train_and_predict(train_x, train.mlb.transform(train_y), test_x,
id2label, args.ensemble)
predict = [id2label[x] for x in predict]
with open(out_file, 'w', encoding='utf8') as f:
for it_predict in predict:
f.write("{}\n".format(it_predict))
print("The ensemble result has been written to {}".format(out_file))