def get_recipe(image_bytes): a=[] b=[] c=[] len1=[] len2=[] tensor = get_tensor(image_bytes) greedy = [True, False, False, False] beam = [-1, -1, -1, -1] temperature = 1.0 numgens = len(greedy) for i in range(numgens): with torch.no_grad(): outputs = model.sample(tensor, greedy=greedy[i],temperature=temperature, beam=beam[i], true_ingrs=None) ingr_ids = outputs['ingr_ids'].cpu().numpy() recipe_ids = outputs['recipe_ids'].cpu().numpy() data_dir = 'M:/Final Project/code/inversecooking-master/data' ingrs_vocab = pickle.load(open(os.path.join(data_dir, 'ingr_vocab.pkl'), 'rb')) vocab = pickle.load(open(os.path.join(data_dir, 'instr_vocab.pkl'), 'rb')) outs, valid = prepare_output(recipe_ids[0], ingr_ids[0], ingrs_vocab, vocab) len1.append(len(outs['ingrs'])) len2.append(len(outs['recipe'])) a.append(outs['title']) b.append(outs['ingrs']) c.append(outs['recipe']) return a,b,c,len1,len2
def generate_recipe(image_tensor):
num_valid = 1
recipes = []
for i in range(numgens):
with torch.no_grad():
outputs = model.sample(
image_tensor,
greedy=greedy[i],
temperature=temperature,
beam=beam[i],
true_ingrs=None,
)
ingr_ids = outputs["ingr_ids"].cpu().numpy()
recipe_ids = outputs["recipe_ids"].cpu().numpy()
outs, valid = prepare_output(recipe_ids[0], ingr_ids[0], ingrs_vocab,
vocab)
if valid["is_valid"]:
logger.warning("Recipe succesfully generated!")
num_valid += 1
# outs['title'], outs['ingrs'], outs['recipe']
logger.warning(outs["title"])
logger.warning(outs["ingrs"])
logger.warning(outs["recipe"])
logger.warning("Generating recipe # {}".format(len(recipes) + 1))
recipes.append(outs)
else:
logger.error("Recipe not valid, stopping...")
if num_valid == 3:
return recipes
return recipes
def getReceipe(img_file, receipeModel):
transf_list_batch = []
transf_list_batch.append(transforms.ToTensor())
transf_list_batch.append(
transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225)))
to_input_transf = transforms.Compose(transf_list_batch)
#greedy = [True, False, False, False]
greedy = [True]
beam = [-1, -1, -1, -1]
temperature = 1.0
numgens = len(greedy)
image_path = img_file
image = Image.open(image_path).convert('RGB')
transf_list = []
transf_list.append(transforms.Resize(256))
transf_list.append(transforms.CenterCrop(224))
transform = transforms.Compose(transf_list)
image_transf = transform(image)
image_tensor = to_input_transf(image_transf).unsqueeze(0).to(device)
#plt.imshow(image_transf)
# plt.axis('off')
# plt.show()
# plt.close()
num_valid = 1
for i in range(numgens):
with torch.no_grad():
outputs = receipeModel.sample(image_tensor,
greedy=greedy[i],
temperature=temperature,
beam=beam[i],
true_ingrs=None)
ingr_ids = outputs['ingr_ids'].cpu().numpy()
recipe_ids = outputs['recipe_ids'].cpu().numpy()
ingrs_vocab = pickle.load(
open(os.path.join(data_dir, 'ingr_vocab.pkl'), 'rb'))
vocab = pickle.load(
open(os.path.join(data_dir, 'instr_vocab.pkl'), 'rb'))
outs, valid = prepare_output(recipe_ids[0], ingr_ids[0], ingrs_vocab,
vocab)
print('-----------------')
print(outs)
print('-----------------')
return outs
return ""
for steps in range(length_steps):
image_step = img_inputs[:, steps, :, :, :]
caption_step = captions[:, steps, :]
ingrs_step = ingr_gt[:, steps, :]
action_step = action_gt[:, steps, :]
true_caps = caption_step.clone()[:, 1:].contiguous()
#new_img_PIL = transforms.ToPILImage()(image_step.cpu().squeeze(0)).convert('RGB')
#new_img_PIL.show() # 处理后的PIL图片
with torch.no_grad():
outputs = model.sample(image_step, true_ingrs=None)
ingr_ids = outputs['ingr_ids'].cpu().numpy()
action_ids = outputs['action_ids'].cpu().numpy()
recipe_ids = outputs['recipe_ids'].cpu().numpy()
outs, valid = prepare_output(recipe_ids[0], ingr_ids[0],
action_ids[0], insts_vocab,
ingrs_vocab, action_vocab)
BOLD = '\033[1m'
END = '\033[0m'
print(BOLD + '\nInstructions:' + END)
print('-' + ' '.join(outs['recipe']))
print(BOLD + '\nIngredients:' + END)
print(', '.join(outs['ingrs']))
print(BOLD + '\nActions:' + END)
print(', '.join(outs['action']))
print('=' * 20)
random.shuffle(demo_imgs)
#demo_urls = ['https://food.fnr.sndimg.com/content/dam/images/food/fullset/2013/12/9/0/FNK_Cheesecake_s4x3.jpg.rend.hgtvcom.826.620.suffix/1387411272847.jpeg'
demo_files =demo_imgs
for img_file in demo_files:
image_path = os.path.join(image_folder, img_file)
image = Image.open(image_path).convert('RGB')
transf_list = []
transf_list.append(transforms.Resize(256))
transf_list.append(transforms.CenterCrop(224))
transform = transforms.Compose(transf_list)
image_transf = transform(image)
image_tensor = to_input_transf(image_transf).unsqueeze(0).to(device)
num_valid = 1
numgens = 1
for i in range(numgens):
with torch.no_grad():
outputs = model.sample(image_tensor, greedy=greedy[i],temperature=temperature, beam=beam[i], true_ingrs=None)
ingr_ids = outputs['ingr_ids'].cpu().numpy()
recipe_ids = outputs['recipe_ids'].cpu().numpy()
outs, valid = prepare_output(recipe_ids[0], ingr_ids[0], ingrs_vocab, vocab)
if True or valid['is_valid']:
print ('RECIPE', num_valid)
print("Image name",img_file)
num_valid+=1
BOLD = '\033[1m'
END = '\033[0m'
#print (BOLD + '\nTitle:' + END,outs['title'])
print (BOLD + '\nIngredients:'+ END)
print (', '.join(outs['ingrs']))
print("Hello")
def process_url(url, food_processor):
try:
response = requests.get(url, stream=True)
image = Image.open(BytesIO(response.content))
print("Got Image")
image_transf = food_processor.transform(image)
image_tensor = food_processor.to_input_transf(image_transf).unsqueeze(
0).to(food_processor.device)
print("Image Transformed")
with torch.no_grad():
outputs = food_processor.model.sample(image_tensor,
greedy=True,
temperature=1.0,
beam=-1,
true_ingrs=None)
print("Got outputs")
ingr_ids = outputs["ingr_ids"].cpu().numpy()
recipe_ids = outputs["recipe_ids"].cpu().numpy()
outs, valid = prepare_output(recipe_ids[0], ingr_ids[0],
food_processor.ingrs_vocab,
food_processor.vocab)
is_valid = valid["is_valid"]
title = outs["title"]
recipes = outs["recipe"]
ingrids = []
new_ingrids = []
alt_ingrids = {}
for ingr in outs["ingrs"]:
ingrids.append((ingr, food_processor.ingr_co2[ingr]))
if len(food_processor.ingr_alternatives[ingr]) > 0:
alt_ingrids[ingr] = food_processor.ingr_alternatives[ingr]
# print(ingr, ingr_alternatives[ingr])
new_ingr = food_processor.ingr_alternatives[ingr][0][0]
new_ingrids.append(
(new_ingr, food_processor.ingr_co2[new_ingr]))
for i, step_ in enumerate(recipes):
step_ = step_.replace(ingr, new_ingr)
recipes[i] = step_
else:
new_ingrids.append((ingr, food_processor.ingr_co2[ingr]))
# ingrids = outs["ingrs"]
except Exception:
is_valid = False
title = "Awesome Recipe"
ingrids = [("Potatoes", 1.0), ("And more Potatoes", 1.0)]
alt_ingrids = {"Potatoes": [("Carrott", 0.5)]}
new_ingrids = [("Carrot", 0.5), ("And more Potatoes", 1.0)]
recipes = ["Eat", "Sleep", "Train", "Repeat"]
traceback.print_stack()
ret_dict = {
"title": title,
"ingredients": ingrids,
"new": new_ingrids,
"alternatives": alt_ingrids,
"instructions": recipes,
"is_valid": is_valid,
"url": url,
}
id_ = get_random_string(16)
req_store[id_] = ret_dict
with open(tmp_json, "w") as fp_:
json.dump(req_store, fp_)
return {"id": id_}
def main(dir_file, image_folder, demo_path, lights):
use_gpu = True
device = torch.device(
'cuda' if torch.cuda.is_available() and use_gpu else 'cpu')
map_loc = None if torch.cuda.is_available() and use_gpu else 'cpu'
ingrs_vocab = pickle.load(
open(os.path.join(dir_file, 'recipe1m_vocab_unit.pkl'), 'rb'))
ingr_vocab_size = len(ingrs_vocab)
t = time.time()
import sys
sys.argv = ['']
del sys
args = get_parser()
args.maxseqlen = 15
args.ingrs_only = False
model = get_model(args, ingr_vocab_size)
# Load the trained model parameters
model_dir = '/home/r8v10/git/InvCo/dataset/new_model/inversecooking/model/checkpoints'
# model_dir = F'{ROOT_DIR}/dataset/model/inversecooking/model/checkpoints'
model_path = os.path.join(model_dir, 'modelbest.ckpt')
model.load_state_dict(torch.load(model_path, map_location=map_loc))
model.to(device)
model.eval()
model.ingrs_only = False
model.recipe_only = False
print('loaded model')
print("Elapsed time:", time.time() - t)
transf_list_batch = []
transf_list_batch.append(transforms.ToTensor())
transf_list_batch.append(
transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225)))
to_input_transf = transforms.Compose(transf_list_batch)
# set to true to load images from demo_urls instead of those in test_imgs folder
use_urls = False
#if True, it will show the recipe even if it's not valid
show_anyways = True
if use_urls:
response = requests.get(demo_path)
image = Image.open(BytesIO(response.content))
else:
image_path = os.path.join(image_folder, demo_path)
image = Image.open(image_path).convert('RGB')
# print('Data path:', image_path)
transf_list = []
transf_list.append(transforms.Resize(256))
transf_list.append(transforms.CenterCrop(224))
transform = transforms.Compose(transf_list)
image_transf = transform(image)
image_tensor = to_input_transf(image_transf).unsqueeze(0).to(device)
num_valid = 1
temperature = 1.0
# greedy = [True, False, False, False]
# beam = [-1, -1, -1, -1]
while True:
with torch.no_grad():
outputs = model.sample(image_tensor,
greedy=False,
temperature=temperature,
beam=-1,
true_ingrs=None)
recipe_ids = outputs['recipe_ids'].cpu().numpy()
outs, valid = prepare_output(recipe_ids[0], ingrs_vocab)
num_valid += 1
if valid['is_valid'] or show_anyways:
if valid['reason'] == 'All ok.':
# print ('RECIPE', num_valid)
# BOLD = '\033[1m'
# END = '\033[0m'
# print (BOLD + '\nTitle:' + END,outs['title'])
# print (BOLD + '\nInstructions:'+END)
# print ('-'+'\n-'.join(outs['recipe']))
# print ('='*20)
#print ("Reason: ", valid['reason'])
break
recommend_id, recommend_lights = search(dir_file, outs['recipe'], lights)
recommend_title, recommend_url = get_recipe(dir_file, recommend_id)
# print('Recommendation of recipe:', recommend_id)
# print('Title:', recommend_title)
# print('Lights:', recommend_lights)
# print('url:', recommend_url)
return outs['title'], outs[
'recipe'], recommend_lights, recommend_title, recommend_url