File size: 1,898 Bytes
7e7a152 | 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 | """CPU checks using synthetic inputs only; no checkpoints or benchmark data."""
import sys,unittest
from pathlib import Path
R=Path(__file__).resolve().parents[1];sys.path[:0]=[str(R),str(R/'tools')]
import numpy as np
import torch
from omini.data.staqa_preprocessing import prepare_staqa_waveform
from omini.models.st_audio_encoder import STAudioEncoder
from hvs_core import perspective,circular_delta
from sound_search_core import score
class NumericChecks(unittest.TestCase):
def test_waveform_normalization_and_resampling(self):
t=np.arange(160000,dtype=np.float32)/16000
wave=np.stack([.1*np.sin(2*np.pi*440*t),.3*np.cos(2*np.pi*220*t)],1)
x=prepare_staqa_waveform(wave,16000)
self.assertEqual(tuple(x.shape),(2,320000));self.assertTrue(torch.isfinite(x).all())
np.testing.assert_allclose(x.square().mean(1).sqrt().numpy(),10**(-14/20),rtol=1e-4)
self.assertFalse(torch.equal(x[0],x[1]))
self.assertEqual(torch.count_nonzero(prepare_staqa_waveform(np.zeros((12,2)),32000)),0)
def test_projection(self):
pano=np.full((512,2048,3),80,dtype=np.uint8)
view=perspective(pano,0,0)
self.assertEqual(view.shape,(360,640,3));self.assertEqual(circular_delta(5,355),10)
def test_sound_submission_gate(self):
label={'target_index':0,'candidates':[{'center':[0,0],'bbox':[990,230,1058,282]},{'center':[120,0],'bbox':[1600,230,1668,282]}]}
self.assertFalse(score([0,0],label,False)['success'])
self.assertFalse(score([120,0],label,True)['success'])
def test_encoder_token_shape(self):
torch.set_num_threads(2)
model=STAudioEncoder(audio_frontend='binaural',num_classes=355,embed_dim=768,depth=12,num_heads=12,audio_bins=40,temporal_layers=2).eval()
with torch.inference_mode():out=model(torch.zeros(1,2,320000))
self.assertEqual(tuple(out.audio_tokens.shape),(1,41,768));self.assertTrue(torch.isfinite(out.audio_tokens).all())
if __name__=='__main__':unittest.main()
|