def __init__(self):
self.x = "hello"
path = untar_data(URLs.PETS)
path_anno = path / 'annotations'
path_images = path / 'images'
fnames = get_image_files(path_images)
np.random.seed(2)
pat = re.compile(r'/([^/]+)_\d+.jpg$')
bs = 6
# create data loaderi
self.data = ImageDataBunch.from_name_re(
path, fnames, pat, ds_tfms=get_transforms(), size=224,
bs=bs).normalize(imagenet_stats)
self.learner = create_cnn(self.dataloader,
models.resnet34,
metrics=error_rate)
cat_images_path = Path("/tmp")
cat_fnames = [
"/{}_1.jpg".format(c)
for c in [
"esdoorn",
"haagbeuk",
"kastanje",
"beuk",
"populier",
"eik",
]
]
cat_data = ImageDataBunch.from_name_re(
cat_images_path,
cat_fnames,
r"/([^/]+)_\d+.jpg$",
ds_tfms=get_transforms(),
size=224,
)
cat_learner = ConvLearner(cat_data, models.resnet50)
cat_learner.model.load_state_dict(
torch.load("rn50-stage-2.pth", map_location="cpu")
)
@app.route("/upload", methods=["POST"])
async def upload(request):
data = await request.form()
bytes = await (data["file"].read())
return predict_image_from_bytes(bytes)
ROOT_DIR = (Path().parent / 'templates').as_posix()
MODEL_NAME = os.environ.get('MODEL_NAME', 'resnet50')
SLOW_PREDICTION = False
app = Starlette()
app.debug = True
app.mount('/static', StaticFiles(directory='static'))
env = Environment(loader=FileSystemLoader(ROOT_DIR), trim_blocks=True)
categories = [
f'/{name.strip()}_1.jpg' for name in Path('categories.txt').open()
]
placeholder_data = ImageDataBunch.from_name_re(Path('/tmp'),
categories,
pat=r'/([^/]+)_\d+.jpg$',
ds_tfms=get_transforms(),
device='cpu',
size=224)
learn = create_cnn(placeholder_data, resnet50)
state = torch.load(f'models/{MODEL_NAME}.pth', map_location='cpu')
learn.model.load_state_dict(state, state)
predictor = Predictor(learn, *imagenet_stats)
@app.route('/')
def home(request):
template = env.get_template('index.html')
return HTMLResponse(template.render(static_url='/static'))