File size: 3,950 Bytes
ea5ea77
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""Measure current video-only DA3 postprocessing on a real pilot source."""
import sys,json,time,subprocess,tempfile,shutil,os
from pathlib import Path
sys.argv=['benchmark','depth']
exec((Path(__file__).parent/'workers.py').read_text().split('def depth(job):')[0])
import requests
scratch=Path(tempfile.mkdtemp(prefix='styleedit-encoding-',dir='/opt'))
try:
 row=json.loads(Path('/workspace/styleedit-streaming-20261024/metadata.jsonl').read_text().splitlines()[0])
 r=requests.get(row['source_video'],timeout=120);r.raise_for_status();src=scratch/'source.mp4';src.write_bytes(r.content)
 meta=ffprobe(src);w,h=meta['width'],meta['height'];n=meta['frames']
 rgb=np.frombuffer(subprocess.check_output(['ffmpeg','-v','error','-threads','2','-i',str(src),'-f','rawvideo','-pix_fmt','rgb24','pipe:1']),np.uint8).reshape(-1,h,w,3)
 ts=time.monotonic()
 with torch.inference_mode():p=model.inference(image=[Image.fromarray(x) for x in rgb],process_res=504,process_res_method='upper_bound_resize',export_dir=None)
 torch.cuda.synchronize();inference=time.monotonic()-ts;dep=p.depth.astype(np.float32)
 ts=time.monotonic();lo,hi=np.percentile(1/dep,[.5,99.5]);percentile=time.monotonic()-ts
 configs=[('current_fast_2',['-c:v','libx264','-threads','2','-preset','fast','-crf','16']),('veryfast_8',['-c:v','libx264','-threads','8','-preset','veryfast','-crf','16']),('nvenc',['-c:v','h264_nvenc','-preset','p4','-rc','constqp','-qp','16'])]
 results=[]
 for repeat in range(2):
  for label,opts in configs:
   path=scratch/(label+'.mp4');total=time.monotonic();convert=write=0;refs=[]
   proc=subprocess.Popen(['ffmpeg','-y','-v','error','-threads','2','-f','rawvideo','-pix_fmt','gray','-s',f'{w}x{h}','-r',meta['avg_frame_rate'],'-i','pipe:0','-an',*opts,'-pix_fmt','yuv420p','-movflags','+faststart',str(path)],stdin=subprocess.PIPE,stderr=subprocess.PIPE)
   try:
    for z in dep:
     t=time.monotonic();z=F.interpolate(torch.from_numpy(z)[None,None],size=(h,w),mode='bilinear',align_corners=False)[0,0].numpy();gray=np.rint(np.clip((1/np.maximum(z,1e-6)-lo)/(hi-lo),0,1)*255).astype(np.uint8);convert+=time.monotonic()-t
     if repeat==0:refs.append(gray)
     t=time.monotonic();proc.stdin.write(gray.tobytes());write+=time.monotonic()-t
    proc.stdin.close();t=time.monotonic();err=proc.stderr.read().decode();rc=proc.wait();wait=time.monotonic()-t
    if rc:raise RuntimeError(err)
    t=time.monotonic();check=ffprobe(path);validate=time.monotonic()-t
    assert check['frames']==n and np.max(np.abs(np.array(check['pts'])-np.array(meta['pts'])))<=.001
    result={'name':label,'repeat':repeat,'conversion_seconds':convert,'pipe_write_seconds':write,'encoder_finish_seconds':wait,'timestamp_validation_seconds':validate,'postprocess_seconds':time.monotonic()-total+percentile,'bytes':path.stat().st_size}
    if repeat==0:
     dec=subprocess.Popen(['ffmpeg','-v','error','-threads','2','-i',str(path),'-f','rawvideo','-pix_fmt','gray','pipe:1'],stdout=subprocess.PIPE);mae=[];mse=[]
     for ref in refs:
      raw=dec.stdout.read(w*h);assert len(raw)==w*h;diff=np.frombuffer(raw,np.uint8).reshape(h,w).astype(np.float32)-ref;mae.append(float(np.abs(diff).mean()));mse.append(float((diff**2).mean()))
     assert dec.wait()==0;result.update(decoded_mae_8bit=float(np.mean(mae)),decoded_rmse_8bit=float(np.sqrt(np.mean(mse))))
    results.append(result);print('ENCODING_RESULT',json.dumps(result),file=sys.stderr,flush=True)
   except Exception as e:
    try:proc.stdin.close()
    except:pass
    proc.wait();results.append({'name':label,'repeat':repeat,'error':str(e)});print('ENCODING_ERROR',label,str(e),file=sys.stderr,flush=True)
 report={'source_id':row['id'],'frames':n,'width':w,'height':h,'inference_seconds':inference,'percentile_seconds':percentile,'npz_saved':False,'results':results}
 atomic(Path(__file__).parent/'encoding-benchmark.json',report)
 print(json.dumps(report,indent=2),file=sys.stderr)
finally:shutil.rmtree(scratch)