Spaces:
Running
on
CPU Upgrade
Running
on
CPU Upgrade
Tuan Tran
commited on
Commit
·
9e6ac41
1
Parent(s):
ab4ab85
update helper functions
Browse files- backend/app.py +15 -26
- backend/config.py +72 -86
- backend/examples.py +0 -1
backend/app.py
CHANGED
|
@@ -1,33 +1,22 @@
|
|
| 1 |
-
from backend.config import (
|
| 2 |
-
ABS_DATASET_DOMAIN,
|
| 3 |
-
get_dataset_config,
|
| 4 |
-
get_datasets,
|
| 5 |
-
)
|
| 6 |
-
from backend.descriptions import (
|
| 7 |
-
DATASET_DESCRIPTIONS,
|
| 8 |
-
DESCRIPTIONS,
|
| 9 |
-
METRIC_DESCRIPTIONS,
|
| 10 |
-
MODEL_DESCRIPTIONS,
|
| 11 |
-
)
|
| 12 |
-
from backend.examples import (
|
| 13 |
-
get_examples_tab,
|
| 14 |
-
)
|
| 15 |
-
from flask import Flask, Response, send_from_directory, request
|
| 16 |
-
from flask_cors import CORS
|
| 17 |
-
import os
|
| 18 |
-
import logging
|
| 19 |
-
import pandas as pd
|
| 20 |
import json
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
get_old_format_dataframe,
|
| 25 |
-
) # Import your function
|
| 26 |
import typing as tp
|
| 27 |
-
import
|
| 28 |
from urllib.parse import unquote
|
| 29 |
-
import mimetypes
|
| 30 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 31 |
|
| 32 |
logger = logging.getLogger(__name__)
|
| 33 |
if not logger.hasHandlers():
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import json
|
| 2 |
+
import logging
|
| 3 |
+
import mimetypes
|
| 4 |
+
import os
|
|
|
|
|
|
|
| 5 |
import typing as tp
|
| 6 |
+
from io import StringIO
|
| 7 |
from urllib.parse import unquote
|
|
|
|
| 8 |
|
| 9 |
+
import pandas as pd
|
| 10 |
+
import requests
|
| 11 |
+
from flask import Flask, Response, request, send_from_directory
|
| 12 |
+
from flask_cors import CORS
|
| 13 |
+
from tools import (get_leaderboard_filters, # Import your function
|
| 14 |
+
get_old_format_dataframe)
|
| 15 |
+
|
| 16 |
+
from backend.config import ABS_DATASET_DOMAIN, get_dataset_config, get_datasets
|
| 17 |
+
from backend.descriptions import (DATASET_DESCRIPTIONS, DESCRIPTIONS,
|
| 18 |
+
METRIC_DESCRIPTIONS, MODEL_DESCRIPTIONS)
|
| 19 |
+
from backend.examples import get_examples_tab
|
| 20 |
|
| 21 |
logger = logging.getLogger(__name__)
|
| 22 |
if not logger.hasHandlers():
|
backend/config.py
CHANGED
|
@@ -1,5 +1,10 @@
|
|
| 1 |
# Change these values to match your dataset structure if loading locally or from a different source.
|
| 2 |
# IMPORTANT: When running from docker more setup is required (e.g. on Huggingface)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
ABS_DATASET_DOMAIN = "https://dl.fbaipublicfiles.com"
|
| 4 |
|
| 5 |
# Sample dataset domain and path for local loading
|
|
@@ -8,10 +13,8 @@ ABS_DATASET_DOMAIN = "https://dl.fbaipublicfiles.com"
|
|
| 8 |
|
| 9 |
ABS_DATASET_PATH = f"{ABS_DATASET_DOMAIN}/omnisealbench/"
|
| 10 |
|
| 11 |
-
|
| 12 |
-
"
|
| 13 |
-
"type": "audio",
|
| 14 |
-
"path": ABS_DATASET_PATH,
|
| 15 |
"first_cols": [
|
| 16 |
"snr",
|
| 17 |
"sisnr",
|
|
@@ -49,79 +52,7 @@ DATASET_CONFIGS = {
|
|
| 49 |
"shush",
|
| 50 |
],
|
| 51 |
},
|
| 52 |
-
"
|
| 53 |
-
"type": "audio",
|
| 54 |
-
"path": ABS_DATASET_PATH,
|
| 55 |
-
"first_cols": ["snr", "sisnr", "stoi", "pesq"],
|
| 56 |
-
"attack_scores": ["bit_acc", "log10_p_value", "TPR", "FPR"],
|
| 57 |
-
"categories": {
|
| 58 |
-
"speed": "Time",
|
| 59 |
-
"updownresample": "Time",
|
| 60 |
-
"echo": "Time",
|
| 61 |
-
"random_noise": "Amplitude",
|
| 62 |
-
"lowpass_filter": "Amplitude",
|
| 63 |
-
"highpass_filter": "Amplitude",
|
| 64 |
-
"bandpass_filter": "Amplitude",
|
| 65 |
-
"smooth": "Amplitude",
|
| 66 |
-
"boost_audio": "Amplitude",
|
| 67 |
-
"duck_audio": "Amplitude",
|
| 68 |
-
"shush": "Amplitude",
|
| 69 |
-
"pink_noise": "Amplitude",
|
| 70 |
-
"aac_compression": "Compression",
|
| 71 |
-
"mp3_compression": "Compression",
|
| 72 |
-
},
|
| 73 |
-
"attacks_with_variations": [
|
| 74 |
-
"random_noise",
|
| 75 |
-
"lowpass_filter",
|
| 76 |
-
"highpass_filter",
|
| 77 |
-
"boost_audio",
|
| 78 |
-
"duck_audio",
|
| 79 |
-
"shush",
|
| 80 |
-
],
|
| 81 |
-
},
|
| 82 |
-
"val2014_1k_v2/image": {
|
| 83 |
-
"type": "image",
|
| 84 |
-
"path": ABS_DATASET_PATH,
|
| 85 |
-
"first_cols": ["psnr", "ssim", "lpips", "decoder_time"],
|
| 86 |
-
"attack_scores": ["bit_acc", "log10_p_value", "TPR", "FPR"],
|
| 87 |
-
"categories": {
|
| 88 |
-
"proportion": "Geometric",
|
| 89 |
-
"collage": "Inpainting",
|
| 90 |
-
"center_crop": "Geometric",
|
| 91 |
-
"rotate": "Geometric",
|
| 92 |
-
"jpeg": "Compression",
|
| 93 |
-
"brightness": "Visual",
|
| 94 |
-
"contrast": "Visual",
|
| 95 |
-
"saturation": "Visual",
|
| 96 |
-
"sharpness": "Visual",
|
| 97 |
-
"resize": "Geometric",
|
| 98 |
-
"overlay_text": "Inpainting",
|
| 99 |
-
"hflip": "Geometric",
|
| 100 |
-
"perspective": "Geometric",
|
| 101 |
-
"median_filter": "Visual",
|
| 102 |
-
"hue": "Visual",
|
| 103 |
-
"gaussian_blur": "Visual",
|
| 104 |
-
"comb": "Mixed",
|
| 105 |
-
"avg": "Averages",
|
| 106 |
-
"none": "Baseline",
|
| 107 |
-
},
|
| 108 |
-
"attacks_with_variations": [
|
| 109 |
-
"center_crop",
|
| 110 |
-
"jpeg",
|
| 111 |
-
"brightness",
|
| 112 |
-
"contrast",
|
| 113 |
-
"saturation",
|
| 114 |
-
"sharpness",
|
| 115 |
-
"resize",
|
| 116 |
-
"perspective",
|
| 117 |
-
"median_filter",
|
| 118 |
-
"hue",
|
| 119 |
-
"gaussian_blur",
|
| 120 |
-
],
|
| 121 |
-
},
|
| 122 |
-
"sa_1b_val_1k/image": {
|
| 123 |
-
"type": "image",
|
| 124 |
-
"path": ABS_DATASET_PATH,
|
| 125 |
"first_cols": ["psnr", "ssim", "lpips", "decoder_time"],
|
| 126 |
"attack_scores": ["bit_acc", "log10_p_value", "TPR", "FPR"],
|
| 127 |
"categories": {
|
|
@@ -159,9 +90,7 @@ DATASET_CONFIGS = {
|
|
| 159 |
"gaussian_blur",
|
| 160 |
],
|
| 161 |
},
|
| 162 |
-
"
|
| 163 |
-
"type": "video",
|
| 164 |
-
"path": ABS_DATASET_PATH,
|
| 165 |
"first_cols": ["psnr", "ssim", "msssim", "lpips", "vmaf", "decoder_time"],
|
| 166 |
"attack_scores": ["bit_acc", "log10_p_value", "TPR", "FPR"],
|
| 167 |
"categories": {
|
|
@@ -198,12 +127,20 @@ DATASET_CONFIGS = {
|
|
| 198 |
"H264rgb",
|
| 199 |
"H265",
|
| 200 |
],
|
| 201 |
-
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 202 |
}
|
| 203 |
|
|
|
|
| 204 |
EXAMPLE_CONFIGS = {
|
| 205 |
"audio": {
|
| 206 |
-
"type": "audio",
|
| 207 |
"dataset_name": "voxpopuli_1k",
|
| 208 |
"path": ABS_DATASET_PATH,
|
| 209 |
"db_key": "voxpopuli",
|
|
@@ -226,24 +163,73 @@ EXAMPLE_CONFIGS = {
|
|
| 226 |
}
|
| 227 |
|
| 228 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 229 |
def get_datasets():
|
| 230 |
grouped = {"audio": [], "image": [], "video": []}
|
| 231 |
for name, cfg in DATASET_CONFIGS.items():
|
| 232 |
dtype = cfg.get("type")
|
| 233 |
if dtype in grouped:
|
| 234 |
grouped[dtype].append(name)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 235 |
return grouped
|
| 236 |
|
| 237 |
|
| 238 |
def get_example_config(type):
|
| 239 |
-
if type in EXAMPLE_CONFIGS:
|
| 240 |
-
return EXAMPLE_CONFIGS[type]
|
| 241 |
-
else:
|
| 242 |
raise ValueError(f"Unknown example type: {type}")
|
| 243 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 244 |
|
| 245 |
def get_dataset_config(dataset_name):
|
| 246 |
if dataset_name in DATASET_CONFIGS:
|
| 247 |
-
|
|
|
|
|
|
|
|
|
|
| 248 |
else:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 249 |
raise ValueError(f"Unknown dataset: {dataset_name}")
|
|
|
|
| 1 |
# Change these values to match your dataset structure if loading locally or from a different source.
|
| 2 |
# IMPORTANT: When running from docker more setup is required (e.g. on Huggingface)
|
| 3 |
+
import os
|
| 4 |
+
from collections import defaultdict
|
| 5 |
+
from copy import deepcopy
|
| 6 |
+
from typing import Any, Dict
|
| 7 |
+
|
| 8 |
ABS_DATASET_DOMAIN = "https://dl.fbaipublicfiles.com"
|
| 9 |
|
| 10 |
# Sample dataset domain and path for local loading
|
|
|
|
| 13 |
|
| 14 |
ABS_DATASET_PATH = f"{ABS_DATASET_DOMAIN}/omnisealbench/"
|
| 15 |
|
| 16 |
+
MODALITY_CONFIG_CONSTANTS = {
|
| 17 |
+
"audio": {
|
|
|
|
|
|
|
| 18 |
"first_cols": [
|
| 19 |
"snr",
|
| 20 |
"sisnr",
|
|
|
|
| 52 |
"shush",
|
| 53 |
],
|
| 54 |
},
|
| 55 |
+
"image": {
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 56 |
"first_cols": ["psnr", "ssim", "lpips", "decoder_time"],
|
| 57 |
"attack_scores": ["bit_acc", "log10_p_value", "TPR", "FPR"],
|
| 58 |
"categories": {
|
|
|
|
| 90 |
"gaussian_blur",
|
| 91 |
],
|
| 92 |
},
|
| 93 |
+
"video": {
|
|
|
|
|
|
|
| 94 |
"first_cols": ["psnr", "ssim", "msssim", "lpips", "vmaf", "decoder_time"],
|
| 95 |
"attack_scores": ["bit_acc", "log10_p_value", "TPR", "FPR"],
|
| 96 |
"categories": {
|
|
|
|
| 127 |
"H264rgb",
|
| 128 |
"H265",
|
| 129 |
],
|
| 130 |
+
}
|
| 131 |
+
}
|
| 132 |
+
|
| 133 |
+
DATASET_CONFIGS = {
|
| 134 |
+
"voxpopuli_1k/audio": {"type": "audio", "path": ABS_DATASET_PATH},
|
| 135 |
+
"ravdess_1k/audio": {"type": "audio", "path": ABS_DATASET_PATH},
|
| 136 |
+
"val2014_1k_v2/image": {"type": "image", "path": ABS_DATASET_PATH},
|
| 137 |
+
"sa_1b_val_1k/image": {"type": "image", "path": ABS_DATASET_PATH},
|
| 138 |
+
"sav_val_full_v2/video": {"type": "video", "path": ABS_DATASET_PATH},
|
| 139 |
}
|
| 140 |
|
| 141 |
+
|
| 142 |
EXAMPLE_CONFIGS = {
|
| 143 |
"audio": {
|
|
|
|
| 144 |
"dataset_name": "voxpopuli_1k",
|
| 145 |
"path": ABS_DATASET_PATH,
|
| 146 |
"db_key": "voxpopuli",
|
|
|
|
| 163 |
}
|
| 164 |
|
| 165 |
|
| 166 |
+
def get_user_dataset():
|
| 167 |
+
datasets = defaultdict(list)
|
| 168 |
+
user_data_dir = os.getenv("OMNISEAL_LEADERBOARD_DATA", "./data")
|
| 169 |
+
if user_data_dir:
|
| 170 |
+
for user_data in os.listdir(user_data_dir):
|
| 171 |
+
if not os.path.isdir(os.path.join(user_data_dir, user_data)):
|
| 172 |
+
continue
|
| 173 |
+
user_dtype = os.listdir(os.path.join(user_data_dir, user_data, "examples"))[0]
|
| 174 |
+
datasets[user_dtype].append(user_data + "/" + user_dtype)
|
| 175 |
+
|
| 176 |
+
return datasets
|
| 177 |
+
|
| 178 |
+
|
| 179 |
def get_datasets():
|
| 180 |
grouped = {"audio": [], "image": [], "video": []}
|
| 181 |
for name, cfg in DATASET_CONFIGS.items():
|
| 182 |
dtype = cfg.get("type")
|
| 183 |
if dtype in grouped:
|
| 184 |
grouped[dtype].append(name)
|
| 185 |
+
|
| 186 |
+
# Add user datasets
|
| 187 |
+
user_datasets = get_user_dataset()
|
| 188 |
+
for dtype, user_names in user_datasets.items():
|
| 189 |
+
if dtype in grouped:
|
| 190 |
+
_names = [name for name in user_names if name not in grouped[dtype]]
|
| 191 |
+
grouped[dtype].extend(_names)
|
| 192 |
return grouped
|
| 193 |
|
| 194 |
|
| 195 |
def get_example_config(type):
|
| 196 |
+
if type not in EXAMPLE_CONFIGS:
|
|
|
|
|
|
|
| 197 |
raise ValueError(f"Unknown example type: {type}")
|
| 198 |
|
| 199 |
+
examples_config: Dict[str, Any] = deepcopy(EXAMPLE_CONFIGS[type])
|
| 200 |
+
|
| 201 |
+
user_datasets = get_user_dataset()
|
| 202 |
+
user_data_dir = os.getenv("OMNISEAL_LEADERBOARD_DATA", "./data")
|
| 203 |
+
if len(user_datasets) > 0:
|
| 204 |
+
assert user_data_dir, f"OMNISEAL_LEADERBOARD_DATA is reset during loading the examples for {type}. Please set it correctly"
|
| 205 |
+
for dtype, user_names in user_datasets.items():
|
| 206 |
+
if dtype == type:
|
| 207 |
+
dataset_name = user_names[0].split("/")[0]
|
| 208 |
+
path = user_data_dir + "/"
|
| 209 |
+
examples_config = {
|
| 210 |
+
"dataset_name": dataset_name,
|
| 211 |
+
"path": path,
|
| 212 |
+
"db_key": dataset_name,
|
| 213 |
+
}
|
| 214 |
+
return examples_config
|
| 215 |
+
|
| 216 |
|
| 217 |
def get_dataset_config(dataset_name):
|
| 218 |
if dataset_name in DATASET_CONFIGS:
|
| 219 |
+
cfg = DATASET_CONFIGS[dataset_name]
|
| 220 |
+
extra_cfg = MODALITY_CONFIG_CONSTANTS.get(cfg["type"], {})
|
| 221 |
+
cfg.update(extra_cfg)
|
| 222 |
+
return cfg
|
| 223 |
else:
|
| 224 |
+
modality = dataset_name.split("/")[-1]
|
| 225 |
+
user_dataset = get_user_dataset()
|
| 226 |
+
if dataset_name in user_dataset.get(modality, []):
|
| 227 |
+
cfg = {
|
| 228 |
+
"type": modality,
|
| 229 |
+
"path": os.getenv("OMNISEAL_LEADERBOARD_DATA"),
|
| 230 |
+
}
|
| 231 |
+
extra_cfg = MODALITY_CONFIG_CONSTANTS.get(cfg["type"], {})
|
| 232 |
+
cfg.update(extra_cfg)
|
| 233 |
+
return cfg
|
| 234 |
+
|
| 235 |
raise ValueError(f"Unknown dataset: {dataset_name}")
|
backend/examples.py
CHANGED
|
@@ -1,7 +1,6 @@
|
|
| 1 |
import ast
|
| 2 |
import json
|
| 3 |
import re
|
| 4 |
-
|
| 5 |
from pathlib import Path
|
| 6 |
|
| 7 |
import requests
|
|
|
|
| 1 |
import ast
|
| 2 |
import json
|
| 3 |
import re
|
|
|
|
| 4 |
from pathlib import Path
|
| 5 |
|
| 6 |
import requests
|