gligen / GLIGEN

Open-Set Grounded Text-to-Image Generation
MIT License
1.92k stars 144 forks source link

Bbox format #35

Closed DavidZah closed 1 year ago

DavidZah commented 1 year ago

Hello, I need to ask in what format is the locations in gligen_inference.py? The images are generating completely out of bbox so I'm wondering what it could be? Here is the code snipset THX in advance :) `def gen_location(): min_size = 0.075

x1 = round(random.uniform(0, 1 - min_size), 8)
x2 = round(random.uniform(x1 + min_size, 1), 8)

y1 = round(random.uniform(0, 1 - min_size), 8)
y2 = round(random.uniform(y1 + min_size, 1), 8)

return [x1, y1, x2, y2]

if name == "main":

parser = argparse.ArgumentParser()
parser.add_argument("--folder", type=str,  default="generation_samples_2", help="root folder for >

parser.add_argument("--batch_size", type=int, default=1, help="")
parser.add_argument("--no_plms", action='store_true', help="use DDIM instead. WARNING: I did not >    parser.add_argument("--guidance_scale", type=float,  default=7.5, help="")
parser.add_argument("--negative_prompt", type=str,  default='longbody, lowres, bad anatomy, bad h>    #parser.add_argument("--negative_prompt", type=str,  default=None, help="")
args = parser.parse_args()

phrases = load_phrases('DATA/phrases.txt')
dict_list = []

dict_list = []

for phrase in phrases:
    random_locations = gen_location()  # Generate random locations
    x = dict(
        ckpt="/home/paperspace/GLIGEN/gligen_checkpoints/diffusion_pytorch_model.bin",
        prompt=phrase,
        locations= random_locations,
        phrases = ["strawberry"],
        alpha_type=[0.3, 0.0, 0.7],
        save_folder_name="generation_box_text_v3"
    )
    dict_list.append(x)

`

DavidZah commented 1 year ago

Ok error is in locations= random_locations, fix locations= [random_locations]