def _transform(self, orig:ndarray, tfms:[Transform], model:nn.Module, sz:int):
for tfm in tfms:
orig,_=tfm(orig, False)
_,val_tfms = tfms_from_stats(inception_stats, sz, crop_type=CropType.NO, aug_tfms=[])
val_tfms.tfms = [tfm for tfm in val_tfms.tfms if not isinstance(tfm, NoCrop)]
orig = val_tfms(orig)
return orig
def get_data(train_names,
val_names,
test_names,
target_size,
batch_size,
n_workers=5):
aug_tfms = [
RandomRotate(30, tfm_y=TfmType.NO),
RandomDihedral(tfm_y=TfmType.NO),
RandomLighting(0.05, 0.05, tfm_y=TfmType.NO)
]
# std and var
stats = A([0.08069, 0.05258, 0.05487, 0.08282],
[0.13704, 0.10145, 0.15313, 0.13814])
tfms = tfms_from_stats(stats,
target_size,
crop_type=CropType.NO,
tfm_y=TfmType.NO,
aug_tfms=aug_tfms)
datasets = ImageData.get_ds(
HPADataset,
(train_names[:-(len(train_names) % batch_size)], cfg.train_dir),
(val_names, cfg.train_dir),
tfms,
test=(test_names, cfg.test_dir))
img_ds = ImageData('./',
datasets,
batch_size,
num_workers=n_workers,
classes=None)
return img_ds
def _transform(self, orig: ndarray, sz: int):
for tfm in self.tfms:
orig, _ = tfm(orig, False)
_, val_tfms = tfms_from_stats(inception_stats,
sz,
crop_type=CropType.NO,
aug_tfms=[])
val_tfms.tfms = [
tfm for tfm in val_tfms.tfms
if not (isinstance(tfm, NoCrop) or isinstance(tfm, Scale))
]
orig = val_tfms(orig)
return orig
class SetupModel(object):
model = classification_model()
labels = get_labels(LABELS_PATH)
tfms = tfms_from_stats(STATS, SZ)[-1]
def __init__(self, f):
self.f = f
file_path = f'/tmp/{STATE_DICT_NAME}'
download_file(BUCKET_NAME, STATE_DICT_NAME, file_path)
state_dict = torch.load(file_path,
map_location=lambda storage, loc: storage)
self.model.load_state_dict(state_dict), self.model.eval()
os.remove(file_path)
def __call__(self, *args, **kwargs):
return self.f(*args, **kwargs)
import os, json, traceback
import urllib.parse
import torch
import numpy as np
from lib.models import classification_model
from lib.utils import download_file, get_labels, open_image_url
from fastai.core import A, T, VV_
from fastai.transforms import tfms_from_stats
BUCKET_NAME = os.environ['BUCKET_NAME']
STATE_DICT_NAME = os.environ['STATE_DICT_NAME']
STATS = A(*eval(os.environ['IMAGE_STATS']))
SZ = int(os.environ['IMAGE_SIZE'])
TFMS = tfms_from_stats(STATS, SZ)[-1]
class SetupModel(object):
model = classification_model()
labels = get_labels(os.environ['LABELS_PATH'])
def __init__(self, f):
self.f = f
file_path = f'/tmp/{STATE_DICT_NAME}'
download_file(BUCKET_NAME, STATE_DICT_NAME, file_path)
state_dict = torch.load(file_path,
map_location=lambda storage, loc: storage)
self.model.load_state_dict(state_dict), self.model.eval()
os.remove(file_path)
