File size: 9,206 Bytes
f60a6c1 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 |
import asyncio
from logging import getLogger
from time import time
from typing import Any
from hashlib import sha256
from json import dumps
from random import randint
from pathlib import Path
from aiohttp import ClientResponseError
from numpy import ndarray
from scorevision.utils.settings import get_settings
from scorevision.utils.bittensor_helpers import load_hotkey_keypair
from scorevision.utils.signing import build_validator_query_params
from scorevision.utils.data_models import SVChallenge
from scorevision.utils.async_clients import get_async_client
from scorevision.utils.video_processing import download_video_cached, FrameStore
from scorevision.utils.image_processing import image_to_base64, pil_from_array
from scorevision.chute_template.schemas import SVFrame
from scorevision.chute_template.schemas import TVPredictInput
from scorevision.vlm_pipeline.domain_specific_schemas.challenge_types import (
parse_challenge_type,
ChallengeType,
)
logger = getLogger(__name__)
class ScoreVisionChallengeError(Exception):
pass
async def get_challenge_from_scorevision() -> tuple[SVChallenge, TVPredictInput]:
try:
chal_api = await get_next_challenge()
except ClientResponseError as e:
raise ScoreVisionChallengeError(f"HTTP error while fetching challenge: {e}")
except ScoreVisionChallengeError as e:
raise e
except Exception as e:
raise Exception(f"Unexpected error while fetching challenge: {e}")
payload, frame_numbers, frames, flows, frame_store = (
await prepare_challenge_payload(challenge=chal_api)
)
if not payload:
if frame_store:
frame_store.unlink()
raise ScoreVisionChallengeError("Failed to prepare payload from challenge.")
# SVChallenge
prompt = f"ScoreVision video task {chal_api.get('task_id')}"
meta = payload.meta | {"seed": chal_api.get("seed", 0)}
canonical = {
"env": "SVEnv",
"prompt": prompt,
"extra": {"meta": meta, "n_frames": len(frames)},
}
cid = sha256(
dumps(canonical, sort_keys=True, separators=(",", ":")).encode()
).hexdigest()
challenge = SVChallenge(
env="SVEnv",
payload=payload,
meta=meta,
prompt=prompt,
challenge_id=cid,
frame_numbers=frame_numbers,
frames=frames,
dense_optical_flow_frames=flows,
)
if frame_store:
frame_store.unlink()
return challenge, payload
async def prepare_challenge_payload(
challenge: dict,
batch_size: int = 64,
*,
video_cache: dict[str, Any] | None = None,
) -> tuple[TVPredictInput, list[int], list[ndarray], list[ndarray], FrameStore]:
settings = get_settings()
video_url = challenge.get("video_url") or challenge.get("asset_url")
if not video_url:
raise ScoreVisionChallengeError("Challenge missing video_url/asset_url")
frame_numbers = list(
range(
settings.SCOREVISION_VIDEO_MIN_FRAME_NUMBER,
settings.SCOREVISION_VIDEO_MAX_FRAME_NUMBER,
)
)
# shuffle(frame_numbers)
start_frame_number = randint(
1,
settings.SCOREVISION_VIDEO_MAX_FRAME_NUMBER
- settings.SCOREVISION_VLM_SELECT_N_FRAMES
- 1,
)
selected_frame_numbers = frame_numbers[
start_frame_number : start_frame_number
+ settings.SCOREVISION_VLM_SELECT_N_FRAMES
]
logger.info(f"Selected Frames for Testing: {selected_frame_numbers}")
cached_store: FrameStore | None = None
cached_path: Path | None = None
if video_cache is not None:
cached_store = video_cache.get("store")
cached_path = video_cache.get("path")
if cached_store is None:
video_name, frame_store = await download_video_cached(
url=video_url,
_frame_numbers=selected_frame_numbers,
cached_path=cached_path,
)
if video_cache is not None:
video_cache["store"] = frame_store
video_cache["path"] = frame_store.video_path
else:
frame_store = cached_store
select_frames: list[ndarray] = []
flow_frames: list[ndarray] = []
for fn in selected_frame_numbers:
frame = await asyncio.to_thread(frame_store.get_frame, fn)
select_frames.append(frame)
flow = await asyncio.to_thread(frame_store.get_flow, fn)
flow_frames.append(flow)
logger.info(f"frames {selected_frame_numbers} successful")
if not select_frames:
raise ScoreVisionChallengeError(
"No Frames were successfully extracted from Video"
)
if not flow_frames:
raise ScoreVisionChallengeError(
"No Dense Optical Flows were successfully computed from Video"
)
height, width = select_frames[0].shape[:2]
meta = {
"version": 1,
"width": width or 0,
"height": height or 0,
"fps": int(
challenge.get("fps") or settings.SCOREVISION_VIDEO_FRAMES_PER_SECOND
),
"task_id": challenge.get("task_id"),
"challenge_type": challenge.get("challenge_type"),
}
if "seed" in challenge:
meta["seed"] = challenge["seed"]
meta["batch_size"] = batch_size
meta["n_keypoints"] = (
32 # TODO: update n_keypoints based on challenge type (32 is for football)
)
payload = TVPredictInput(url=video_url, meta=meta)
return (
payload,
selected_frame_numbers,
select_frames,
flow_frames,
frame_store,
)
async def get_next_challenge() -> dict:
"""
Fetches the next video challenge from ScoreVision API.
Returns a dict like:
{
"task_id": "...", # we will propagate this end-to-end
"video_url": "...", # or "asset_url"
"fps": 25|30, # optional (fallback 30)
"seed": <int>, # optional
...
}
"""
settings = get_settings()
if not settings.SCOREVISION_API:
raise ScoreVisionChallengeError("SCOREVISION_API is not set.")
keypair = load_hotkey_keypair(
wallet_name=settings.BITTENSOR_WALLET_COLD,
hotkey_name=settings.BITTENSOR_WALLET_HOT,
)
# Build query parameters required to authenticate with the validator API
params = build_validator_query_params(keypair)
session = await get_async_client()
async with session.get(
f"{settings.SCOREVISION_API}/api/tasks/next/v2", params=params
) as response:
response.raise_for_status()
challenge = await response.json() or None
if not challenge:
raise ScoreVisionChallengeError("No challenge available from API")
if "id" in challenge and "task_id" not in challenge:
challenge["task_id"] = challenge.pop("id")
if not (challenge.get("video_url") or challenge.get("asset_url")):
raise ScoreVisionChallengeError("Challenge missing video url.")
ct = (
parse_challenge_type(challenge.get("challenge_type"))
or ChallengeType.FOOTBALL
)
challenge["challenge_type"] = ct.value
logger.info(f"Fetched challenge: task_id={challenge.get('task_id')}")
return challenge
def build_svchallenge_from_parts(
chal_api: dict,
payload: TVPredictInput,
frame_numbers: list[int],
frames: list[ndarray],
flows: list[ndarray],
) -> SVChallenge:
prompt = f"ScoreVision video task {chal_api.get('task_id')}"
meta = payload.meta | {"seed": chal_api.get("seed", 0)}
canonical = {
"env": "SVEnv",
"prompt": prompt,
"extra": {"meta": meta, "n_frames": len(frames)},
}
cid = sha256(
dumps(canonical, sort_keys=True, separators=(",", ":")).encode()
).hexdigest()
ct = parse_challenge_type(chal_api.get("challenge_type"))
return SVChallenge(
env="SVEnv",
payload=payload,
meta=meta,
prompt=prompt,
challenge_id=cid,
frame_numbers=frame_numbers,
frames=frames,
dense_optical_flow_frames=flows,
api_task_id=chal_api.get("task_id"),
challenge_type=ct,
)
async def get_challenge_from_scorevision_with_source(
*,
video_cache: dict[str, Any] | None = None,
) -> tuple[SVChallenge, TVPredictInput, dict, FrameStore]:
try:
chal_api = await get_next_challenge()
except ClientResponseError as e:
raise ScoreVisionChallengeError(f"HTTP error while fetching challenge: {e}")
except ScoreVisionChallengeError as e:
raise e
except Exception as e:
raise Exception(f"Unexpected error while fetching challenge: {e}")
payload, frame_numbers, frames, flows, frame_store = (
await prepare_challenge_payload(
challenge=chal_api,
video_cache=video_cache,
)
)
if not payload:
raise ScoreVisionChallengeError("Failed to prepare payload from challenge.")
challenge = build_svchallenge_from_parts(
chal_api=chal_api,
payload=payload,
frame_numbers=frame_numbers,
frames=frames,
flows=flows,
)
return challenge, payload, chal_api, frame_store
|