LPX55
commited on
Commit
·
80c3f0c
1
Parent(s):
4cb6734
refactor: ONNX model loading and caching functions
Browse files- app.py +1 -36
- utils/onnx_model_loader.py +33 -0
app.py
CHANGED
@@ -17,6 +17,7 @@ import torch
|
|
17 |
|
18 |
from utils.utils import softmax, augment_image, preprocess_resize_256, preprocess_resize_224, postprocess_pipeline, postprocess_logits, postprocess_binary_output, to_float_scalar, infer_gradio_api, preprocess_gradio_api, postprocess_gradio_api
|
19 |
from utils.onnx_helpers import preprocess_onnx_input, postprocess_onnx_output
|
|
|
20 |
from forensics.gradient import gradient_processing
|
21 |
from forensics.minmax import minmax_process
|
22 |
from forensics.ela import ELA
|
@@ -97,46 +98,10 @@ def register_model_with_metadata(model_id, model, preprocess, postprocess, class
|
|
97 |
MODEL_REGISTRY[model_id] = entry
|
98 |
|
99 |
|
100 |
-
def load_onnx_model_and_preprocessor(hf_model_id):
|
101 |
-
# model_dir = snapshot_download(repo_id=hf_model_id, local_dir_use_symlinks=False)
|
102 |
-
|
103 |
-
# Create a unique local directory for each ONNX model
|
104 |
-
model_specific_dir = os.path.join("./models", hf_model_id.replace('/', '_'))
|
105 |
-
os.makedirs(model_specific_dir, exist_ok=True)
|
106 |
-
|
107 |
-
# Use hf_hub_download to get specific files into the model-specific directory
|
108 |
-
onnx_model_path = hf_hub_download(repo_id=hf_model_id, filename="model_quantized.onnx", subfolder="onnx", local_dir=model_specific_dir, local_dir_use_symlinks=False)
|
109 |
-
|
110 |
-
# Load preprocessor config
|
111 |
-
preprocessor_config = {}
|
112 |
-
try:
|
113 |
-
preprocessor_config_path = hf_hub_download(repo_id=hf_model_id, filename="preprocessor_config.json", local_dir=model_specific_dir, local_dir_use_symlinks=False)
|
114 |
-
with open(preprocessor_config_path, 'r') as f:
|
115 |
-
preprocessor_config = json.load(f)
|
116 |
-
except Exception as e:
|
117 |
-
logger.warning(f"Could not download or load preprocessor_config.json for {hf_model_id}: {e}")
|
118 |
-
|
119 |
-
# Load model config for class names if available
|
120 |
-
model_config = {}
|
121 |
-
try:
|
122 |
-
model_config_path = hf_hub_download(repo_id=hf_model_id, filename="config.json", local_dir=model_specific_dir, local_dir_use_symlinks=False)
|
123 |
-
with open(model_config_path, 'r') as f:
|
124 |
-
model_config = json.load(f)
|
125 |
-
except Exception as e:
|
126 |
-
logger.warning(f"Could not download or load config.json for {hf_model_id}: {e}")
|
127 |
-
|
128 |
-
return onnxruntime.InferenceSession(onnx_model_path), preprocessor_config, model_config
|
129 |
-
|
130 |
|
131 |
# Cache for ONNX sessions and preprocessors
|
132 |
_onnx_model_cache = {}
|
133 |
|
134 |
-
def get_onnx_model_from_cache(hf_model_id):
|
135 |
-
if hf_model_id not in _onnx_model_cache:
|
136 |
-
logger.info(f"Loading ONNX model and preprocessor for {hf_model_id}...")
|
137 |
-
_onnx_model_cache[hf_model_id] = load_onnx_model_and_preprocessor(hf_model_id)
|
138 |
-
return _onnx_model_cache[hf_model_id]
|
139 |
-
|
140 |
|
141 |
def infer_onnx_model(hf_model_id, preprocessed_image_np, model_config: dict):
|
142 |
try:
|
|
|
17 |
|
18 |
from utils.utils import softmax, augment_image, preprocess_resize_256, preprocess_resize_224, postprocess_pipeline, postprocess_logits, postprocess_binary_output, to_float_scalar, infer_gradio_api, preprocess_gradio_api, postprocess_gradio_api
|
19 |
from utils.onnx_helpers import preprocess_onnx_input, postprocess_onnx_output
|
20 |
+
from utils.onnx_model_loader import load_onnx_model_and_preprocessor, get_onnx_model_from_cache
|
21 |
from forensics.gradient import gradient_processing
|
22 |
from forensics.minmax import minmax_process
|
23 |
from forensics.ela import ELA
|
|
|
98 |
MODEL_REGISTRY[model_id] = entry
|
99 |
|
100 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
101 |
|
102 |
# Cache for ONNX sessions and preprocessors
|
103 |
_onnx_model_cache = {}
|
104 |
|
|
|
|
|
|
|
|
|
|
|
|
|
105 |
|
106 |
def infer_onnx_model(hf_model_id, preprocessed_image_np, model_config: dict):
|
107 |
try:
|
utils/onnx_model_loader.py
ADDED
@@ -0,0 +1,33 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import json
|
3 |
+
import onnxruntime
|
4 |
+
from huggingface_hub import hf_hub_download
|
5 |
+
import logging
|
6 |
+
|
7 |
+
def load_onnx_model_and_preprocessor(hf_model_id):
|
8 |
+
model_specific_dir = os.path.join("./models", hf_model_id.replace('/', '_'))
|
9 |
+
os.makedirs(model_specific_dir, exist_ok=True)
|
10 |
+
onnx_model_path = hf_hub_download(repo_id=hf_model_id, filename="model_quantized.onnx", subfolder="onnx", local_dir=model_specific_dir, local_dir_use_symlinks=False)
|
11 |
+
preprocessor_config = {}
|
12 |
+
try:
|
13 |
+
preprocessor_config_path = hf_hub_download(repo_id=hf_model_id, filename="preprocessor_config.json", local_dir=model_specific_dir, local_dir_use_symlinks=False)
|
14 |
+
with open(preprocessor_config_path, 'r') as f:
|
15 |
+
preprocessor_config = json.load(f)
|
16 |
+
except Exception as e:
|
17 |
+
logging.getLogger(__name__).warning(f"Could not download or load preprocessor_config.json for {hf_model_id}: {e}")
|
18 |
+
model_config = {}
|
19 |
+
try:
|
20 |
+
model_config_path = hf_hub_download(repo_id=hf_model_id, filename="config.json", local_dir=model_specific_dir, local_dir_use_symlinks=False)
|
21 |
+
with open(model_config_path, 'r') as f:
|
22 |
+
model_config = json.load(f)
|
23 |
+
except Exception as e:
|
24 |
+
logging.getLogger(__name__).warning(f"Could not download or load config.json for {hf_model_id}: {e}")
|
25 |
+
return onnxruntime.InferenceSession(onnx_model_path), preprocessor_config, model_config
|
26 |
+
|
27 |
+
def get_onnx_model_from_cache(hf_model_id, _onnx_model_cache, load_func):
|
28 |
+
import logging
|
29 |
+
logger = logging.getLogger(__name__)
|
30 |
+
if hf_model_id not in _onnx_model_cache:
|
31 |
+
logger.info(f"Loading ONNX model and preprocessor for {hf_model_id}...")
|
32 |
+
_onnx_model_cache[hf_model_id] = load_func(hf_model_id)
|
33 |
+
return _onnx_model_cache[hf_model_id]
|