11155aead4
This has been a long time coming. Cleans up messy "GT" nomenclature and simplifies ExtensibleTraner.feed_data
58 lines
2.8 KiB
Python
58 lines
2.8 KiB
Python
import os
|
|
|
|
import torch
|
|
import os.path as osp
|
|
import torchvision
|
|
from torch.utils.data import BatchSampler
|
|
|
|
import models.eval.evaluator as evaluator
|
|
from pytorch_fid import fid_score
|
|
|
|
|
|
# Evaluate that feeds a LR structure into the input, then calculates a FID score on the results added to
|
|
# the interpolated LR structure.
|
|
from data.stylegan2_dataset import Stylegan2Dataset
|
|
|
|
|
|
class SrStyleTransferEvaluator(evaluator.Evaluator):
|
|
def __init__(self, model, opt_eval, env):
|
|
super().__init__(model, opt_eval, env)
|
|
self.batches_per_eval = opt_eval['batches_per_eval']
|
|
self.batch_sz = opt_eval['batch_size']
|
|
self.im_sz = opt_eval['image_size']
|
|
self.scale = opt_eval['scale']
|
|
self.fid_real_samples = opt_eval['real_fid_path']
|
|
self.embedding_generator = opt_eval['embedding_generator']
|
|
self.gen_output_index = opt_eval['gen_index'] if 'gen_index' in opt_eval.keys() else 0
|
|
self.dataset = Stylegan2Dataset({'path': self.fid_real_samples,
|
|
'target_size': self.im_sz,
|
|
'aug_prob': 0,
|
|
'transparent': False})
|
|
self.sampler = BatchSampler(self.dataset, self.batch_sz, False)
|
|
|
|
def perform_eval(self):
|
|
embedding_generator = self.env['generators'][self.embedding_generator]
|
|
fid_fake_path = osp.join(self.env['base_path'], "..", "fid_fake", str(self.env["step"]))
|
|
os.makedirs(fid_fake_path, exist_ok=True)
|
|
fid_real_path = osp.join(self.env['base_path'], "..", "fid_real", str(self.env["step"]))
|
|
os.makedirs(fid_real_path, exist_ok=True)
|
|
counter = 0
|
|
for batch in self.sampler:
|
|
noise = torch.FloatTensor(self.batch_sz, 3, self.im_sz, self.im_sz).uniform_(0., 1.).to(self.env['device'])
|
|
batch_hq = [e['hq'] for e in batch]
|
|
batch_hq = torch.stack(batch_hq, dim=0).to(self.env['device'])
|
|
resized_batch = torch.nn.functional.interpolate(batch_hq, scale_factor=1/self.scale, mode="area")
|
|
embedding = embedding_generator(resized_batch)
|
|
gen = self.model(noise, embedding)
|
|
if not isinstance(gen, list) and not isinstance(gen, tuple):
|
|
gen = [gen]
|
|
gen = gen[self.gen_output_index]
|
|
out = gen + torch.nn.functional.interpolate(resized_batch, scale_factor=self.scale, mode='bilinear')
|
|
for b in range(self.batch_sz):
|
|
torchvision.utils.save_image(out[b], osp.join(fid_fake_path, "%i_.png" % (counter)))
|
|
torchvision.utils.save_image(batch_hq[b], osp.join(fid_real_path, "%i_.png" % (counter)))
|
|
counter += 1
|
|
|
|
return {"fid": fid_score.calculate_fid_given_paths([fid_real_path, fid_fake_path], self.batch_sz, True,
|
|
2048)}
|