if index == 10:
warmup_t = time.time()
data_line = data_line.decode('utf-8')
substr = data_line.strip("\n").split("\t")
img_path, label = data_dir + "/" + substr[0], substr[1]
data = {'img_path': img_path, 'label': label}
with open(data['img_path'], 'rb') as f:
img = f.read()
data['image'] = img
st = time.time()
batch = transform(data, ops)
batch_pred = [0] * len(batch)
for i in range(len(batch)):
batch_pred[i] = paddle.to_tensor(
np.expand_dims(batch[i], axis=0))
st = time.time()
node, edge = model(batch_pred)
node = F.softmax(node, -1)
count_t.append(time.time() - st)
draw_kie_result(batch, node, idx_to_cls, index)
logger.info("success!")
logger.info("It took {} s for predict {} images.".format(
np.sum(count_t), len(count_t)))
ips = len(count_t[warmup_times:]) / np.sum(count_t[warmup_times:])
logger.info("The ips is {} images/s".format(ips))
if __name__ == '__main__':
config, device, logger, vdl_writer = program.preprocess()
main()
])
logger.info(
"The sensitivity analysis results of model parameters saved in sen.pickle"
)
# calculate pruned params's ratio
params_sensitive = pruner._get_ratios_by_loss(params_sensitive, loss=0.02)
for key in params_sensitive.keys():
logger.info("{}, {}".format(key, params_sensitive[key]))
#params_sensitive = {}
#for param in model.parameters():
# if 'transpose' not in param.name and 'linear' not in param.name:
# params_sensitive[param.name] = 0.1
plan = pruner.prune_vars(params_sensitive, [0])
flops = paddle.flops(model, [1, 3, 640, 640])
logger.info("FLOPs after pruning: {}".format(flops))
# start train
program.train(config, train_dataloader, valid_dataloader, device, model,
loss_class, optimizer, lr_scheduler, post_process_class,
eval_class, pre_best_model_dict, logger, vdl_writer)
if __name__ == '__main__':
config, device, logger, vdl_writer = program.preprocess(is_train=True)
main(config, device, logger, vdl_writer)
