Spaces:
Runtime error
Runtime error
File size: 5,775 Bytes
e82dff2 3b61cce 7617596 3b61cce ad3ee60 3b61cce a8797f1 bd0d673 3b61cce 0312353 3b61cce 19327c9 3b61cce 0312353 3b61cce 0312353 3b61cce 0312353 3b61cce a8797f1 3b61cce a8797f1 ad3ee60 0312353 6eaf487 7617596 6eaf487 7617596 0312353 6eaf487 7617596 0312353 7617596 3b61cce 7617596 0312353 7617596 0312353 7617596 3b61cce 0312353 7617596 19327c9 ad3ee60 19327c9 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 |
from __future__ import annotations
import os
import pathlib
import shlex
import shutil
import subprocess
import sys
import cv2
import torch
from label_prettify import label_prettify
repo_dir = pathlib.Path(__file__).parent
submodule_dir = repo_dir / 'prismer'
sys.path.insert(0, submodule_dir.as_posix())
from dataset import create_dataset, create_loader
from model.prismer_caption import PrismerCaption
def download_models() -> None:
if not pathlib.Path('prismer/experts/expert_weights/').exists():
subprocess.run(shlex.split('python download_checkpoints.py --download_experts=True'), cwd='prismer')
model_names = [
'vqa_prismer_base',
'vqa_prismer_large',
'pretrain_prismer_base',
'pretrain_prismer_large',
]
for model_name in model_names:
if pathlib.Path(f'prismer/logging/{model_name}').exists():
continue
subprocess.run(shlex.split(f'python download_checkpoints.py --download_models={model_name}'), cwd='prismer')
def build_deformable_conv() -> None:
subprocess.run(shlex.split('sh make.sh'), cwd='prismer/experts/segmentation/mask2former/modeling/pixel_decoder/ops')
def run_experts(image_path: str) -> tuple[str | None, ...]:
helper_dir = submodule_dir / 'helpers'
shutil.rmtree(helper_dir, ignore_errors=True)
image_dir = helper_dir / 'images'
image_dir.mkdir(parents=True, exist_ok=True)
out_path = image_dir / 'image.jpg'
cv2.imwrite(out_path.as_posix(), cv2.imread(image_path))
expert_names = ['depth', 'edge', 'normal', 'objdet', 'ocrdet', 'segmentation']
for expert_name in expert_names:
env = os.environ.copy()
if 'PYTHONPATH' in env:
env['PYTHONPATH'] = f'{submodule_dir.as_posix()}:{env["PYTHONPATH"]}'
else:
env['PYTHONPATH'] = submodule_dir.as_posix()
subprocess.run(shlex.split(f'python experts/generate_{expert_name}.py'), cwd='prismer', env=env, check=True)
keys = ['depth', 'edge', 'normal', 'seg_coco', 'obj_detection', 'ocr_detection']
results = [pathlib.Path('prismer/helpers/labels') / key / 'helpers/images/image.png' for key in keys]
return tuple(path.as_posix() for path in results)
class Model:
def __init__(self):
self.config = None
self.model = None
self.tokenizer = None
self.exp_name = ''
self.mode = ''
def set_model(self, exp_name: str) -> None:
if exp_name == self.exp_name:
return
if self.mode == 'caption':
config = {
'dataset': 'demo',
'data_path': 'prismer/helpers',
'label_path': 'prismer/helpers/labels',
'experts': ['depth', 'normal', 'seg_coco', 'edge', 'obj_detection', 'ocr_detection'],
'image_resolution': 480,
'prismer_model': 'prismer_base' if self.exp_name == 'Prismer-Base' else 'prismer_large',
'freeze': 'freeze_vision',
'prefix': 'A picture of',
}
model = PrismerCaption(config)
state_dict = torch.load(f'prismer/logging/caption_{exp_name}/pytorch_model.bin', map_location='cuda:0')
elif self.mode == 'vqa':
config = {
'dataset': 'demo',
'data_path': 'prismer/helpers',
'label_path': 'prismer/helpers/labels',
'experts': ['depth', 'normal', 'seg_coco', 'edge', 'obj_detection', 'ocr_detection'],
'image_resolution': 480,
'prismer_model': 'prismer_base' if self.exp_name == 'Prismer-Base' else 'prismer_large',
'freeze': 'freeze_vision',
'prefix': 'A picture of',
}
model = PrismerCaption(config)
state_dict = torch.load(f'prismer/logging/caption_{exp_name}/pytorch_model.bin', map_location='cuda:0')
model.load_state_dict(state_dict)
model.eval()
self.config = config
self.model = model
self.tokenizer = model.tokenizer
self.exp_name = exp_name
@torch.inference_mode()
def run_caption_model(self, exp_name: str) -> str:
self.set_model(exp_name)
_, test_dataset = create_dataset('caption', self.config)
test_loader = create_loader(test_dataset, batch_size=1, num_workers=4, train=False)
experts, _ = next(iter(test_loader))
captions = self.model(experts, train=False, prefix=self.config['prefix'])
captions = self.tokenizer(captions, max_length=30, padding='max_length', return_tensors='pt').input_ids
caption = captions.to(experts['rgb'].device)[0]
caption = self.tokenizer.decode(caption, skip_special_tokens=True)
caption = caption.capitalize() + '.'
return caption
def run_caption(self, image_path: str, model_name: str) -> tuple[str | None, ...]:
out_paths = run_experts(image_path)
caption = self.run_caption_model(model_name)
label_prettify(image_path, out_paths)
return caption, *out_paths
@torch.inference_mode()
def run_vqa_model(self, exp_name: str) -> str:
self.set_model(exp_name)
_, test_dataset = create_dataset('vqa', self.config)
test_loader = create_loader(test_dataset, batch_size=1, num_workers=4, train=False)
experts, _ = next(iter(test_loader))
captions = self.model(experts, train=False, prefix=self.config['prefix'])
captions = self.tokenizer(captions, max_length=30, padding='max_length', return_tensors='pt').input_ids
caption = captions.to(experts['rgb'].device)[0]
caption = self.tokenizer.decode(caption, skip_special_tokens=True)
caption = caption.capitalize() + '.'
return caption |