Upload selected files from lensless_exps subfolders
Browse files- 32x32_librispeech_mse_PSF_Unet4M_U5_Unet4M/checkpoint-epoch100.pth +3 -0
- 32x32_librispeech_mse_PSF_Unet4M_U5_Unet4M/config.yaml +324 -0
- 32x32_librispeech_mse_ssim_PSF_Unet4M_U5_Unet4M/checkpoint-epoch100.pth +3 -0
- 32x32_librispeech_mse_ssim_PSF_Unet4M_U5_Unet4M/config.yaml +324 -0
- 32x32_librispeech_mse_ssim_raw_ssim_PSF_Unet4M_U5_Unet4M/checkpoint-epoch100.pth +3 -0
- 32x32_librispeech_mse_ssim_raw_ssim_PSF_Unet4M_U5_Unet4M/config.yaml +315 -0
- 32x32_librispeech_mse_ssim_raw_ssim_l1_PSF_Unet4M_U5_Unet4M/config.yaml +324 -0
- 32x32_librispeech_mse_ssim_raw_ssim_mel_PSF_Unet4M_U5_Unet4M/checkpoint-epoch100.pth +3 -0
- 32x32_librispeech_mse_ssim_raw_ssim_mel_PSF_Unet4M_U5_Unet4M/config.yaml +324 -0
- 32x32_random_mse_ssim_raw_ssim_PSF_Unet4M_U5_Unet4M/checkpoint-epoch100.pth +3 -0
- 32x32_random_mse_ssim_raw_ssim_PSF_Unet4M_U5_Unet4M/config.yaml +314 -0
32x32_librispeech_mse_PSF_Unet4M_U5_Unet4M/checkpoint-epoch100.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2809709a77d519fae6b2ba893d2d09748faf5373a2110cbb688fb1375fff2a6f
|
| 3 |
+
size 97418362
|
32x32_librispeech_mse_PSF_Unet4M_U5_Unet4M/config.yaml
ADDED
|
@@ -0,0 +1,324 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
model:
|
| 2 |
+
_target_: src.model.LenslessWrapper
|
| 3 |
+
use_loader: false
|
| 4 |
+
loader_kwargs: null
|
| 5 |
+
use_batch_video_version: false
|
| 6 |
+
freeze_weights: false
|
| 7 |
+
psf_path: data/digicam_psf/SIM_psf.png
|
| 8 |
+
psf_loader_kwargs:
|
| 9 |
+
downsample: 8
|
| 10 |
+
return_bg: false
|
| 11 |
+
grayscale_psf: true
|
| 12 |
+
recon_name: UnrolledADMM
|
| 13 |
+
recon_kwargs:
|
| 14 |
+
n_iter: 5
|
| 15 |
+
mu1: 0.0001
|
| 16 |
+
mu2: 0.0001
|
| 17 |
+
mu3: 0.0001
|
| 18 |
+
tau: 0.0002
|
| 19 |
+
pre_process:
|
| 20 |
+
_target_: lensless.recon.drunet.network_unet.UNetRes
|
| 21 |
+
in_nc: 2
|
| 22 |
+
out_nc: 1
|
| 23 |
+
nc:
|
| 24 |
+
- 32
|
| 25 |
+
- 64
|
| 26 |
+
- 112
|
| 27 |
+
- 128
|
| 28 |
+
nb: 4
|
| 29 |
+
act_mode: R
|
| 30 |
+
downsample_mode: strideconv
|
| 31 |
+
upsample_mode: convtranspose
|
| 32 |
+
post_process:
|
| 33 |
+
_target_: lensless.recon.drunet.network_unet.UNetRes
|
| 34 |
+
in_nc: 2
|
| 35 |
+
out_nc: 1
|
| 36 |
+
nc:
|
| 37 |
+
- 32
|
| 38 |
+
- 64
|
| 39 |
+
- 116
|
| 40 |
+
- 128
|
| 41 |
+
nb: 4
|
| 42 |
+
act_mode: R
|
| 43 |
+
downsample_mode: strideconv
|
| 44 |
+
upsample_mode: convtranspose
|
| 45 |
+
psf_network:
|
| 46 |
+
_target_: lensless.recon.drunet.network_unet.UNetRes
|
| 47 |
+
in_nc: 2
|
| 48 |
+
out_nc: 1
|
| 49 |
+
nc:
|
| 50 |
+
- 4
|
| 51 |
+
- 8
|
| 52 |
+
- 16
|
| 53 |
+
- 32
|
| 54 |
+
nb: 4
|
| 55 |
+
act_mode: R
|
| 56 |
+
downsample_mode: strideconv
|
| 57 |
+
upsample_mode: convtranspose
|
| 58 |
+
psf_residual: false
|
| 59 |
+
skip_unrolled: false
|
| 60 |
+
return_intermediate: false
|
| 61 |
+
writer:
|
| 62 |
+
_target_: src.logger.WandBWriter
|
| 63 |
+
project_name: lenslessmic
|
| 64 |
+
entity: null
|
| 65 |
+
run_name: 32x32_librispeech_mse_PSF_Unet4M_U5_Unet4M
|
| 66 |
+
mode: online
|
| 67 |
+
loss_names:
|
| 68 |
+
- loss
|
| 69 |
+
- codec_mse_loss
|
| 70 |
+
- codec_ssim_loss
|
| 71 |
+
- raw_codec_ssim_loss
|
| 72 |
+
- raw_codec_l1_loss
|
| 73 |
+
- audio_l1_loss
|
| 74 |
+
- audio_sisdr_loss
|
| 75 |
+
- audio_stft_loss
|
| 76 |
+
- audio_mel_loss
|
| 77 |
+
log_checkpoints: false
|
| 78 |
+
id_length: 8
|
| 79 |
+
names:
|
| 80 |
+
- input-1:frame
|
| 81 |
+
- input-2:frame
|
| 82 |
+
- input-3:frame
|
| 83 |
+
- input-4:frame
|
| 84 |
+
figsize:
|
| 85 |
+
- 15
|
| 86 |
+
- 15
|
| 87 |
+
sample_rate: 16000
|
| 88 |
+
run_id: 1r070ti9
|
| 89 |
+
metrics:
|
| 90 |
+
device: auto
|
| 91 |
+
train:
|
| 92 |
+
- _target_: src.metrics.SISDRMetric
|
| 93 |
+
name: SISDR
|
| 94 |
+
- _target_: src.metrics.PSNRMetric
|
| 95 |
+
name: PSNR
|
| 96 |
+
- _target_: src.metrics.QuantizationMatchMetric
|
| 97 |
+
name: QuantizationMatch-all
|
| 98 |
+
codebook_index: all
|
| 99 |
+
inference:
|
| 100 |
+
- _target_: src.metrics.SISDRMetric
|
| 101 |
+
name: SISDR
|
| 102 |
+
- _target_: src.metrics.STOIMetric
|
| 103 |
+
name: STOI
|
| 104 |
+
- _target_: src.metrics.WERMetric
|
| 105 |
+
name: WER
|
| 106 |
+
- _target_: src.metrics.PESQMetric
|
| 107 |
+
name: PESQ
|
| 108 |
+
- _target_: src.metrics.MelMetric
|
| 109 |
+
name: Mel
|
| 110 |
+
audio_mel_config:
|
| 111 |
+
n_mels:
|
| 112 |
+
- 5
|
| 113 |
+
- 10
|
| 114 |
+
- 20
|
| 115 |
+
- 40
|
| 116 |
+
- 80
|
| 117 |
+
- 160
|
| 118 |
+
- 320
|
| 119 |
+
window_lengths:
|
| 120 |
+
- 32
|
| 121 |
+
- 64
|
| 122 |
+
- 128
|
| 123 |
+
- 256
|
| 124 |
+
- 512
|
| 125 |
+
- 1024
|
| 126 |
+
- 2048
|
| 127 |
+
mel_fmin:
|
| 128 |
+
- 0
|
| 129 |
+
- 0
|
| 130 |
+
- 0
|
| 131 |
+
- 0
|
| 132 |
+
- 0
|
| 133 |
+
- 0
|
| 134 |
+
- 0
|
| 135 |
+
mel_fmax:
|
| 136 |
+
- null
|
| 137 |
+
- null
|
| 138 |
+
- null
|
| 139 |
+
- null
|
| 140 |
+
- null
|
| 141 |
+
- null
|
| 142 |
+
- null
|
| 143 |
+
pow: 1.0
|
| 144 |
+
clamp_eps: 1.0e-05
|
| 145 |
+
mag_weight: 0.0
|
| 146 |
+
- _target_: src.metrics.STFTMetric
|
| 147 |
+
name: STFT
|
| 148 |
+
audio_stft_config:
|
| 149 |
+
window_lengths:
|
| 150 |
+
- 2048
|
| 151 |
+
- 512
|
| 152 |
+
- _target_: src.metrics.QuantizationMatchMetric
|
| 153 |
+
name: QuantizationMatch-all
|
| 154 |
+
codebook_index: all
|
| 155 |
+
- _target_: src.metrics.QuantizationMatchMetric
|
| 156 |
+
name: QuantizationMatch-1
|
| 157 |
+
codebook_index: 1
|
| 158 |
+
- _target_: src.metrics.QuantizationMatchMetric
|
| 159 |
+
name: QuantizationMatch-2
|
| 160 |
+
codebook_index: 2
|
| 161 |
+
- _target_: src.metrics.PSNRMetric
|
| 162 |
+
name: PSNR
|
| 163 |
+
- _target_: src.metrics.SSIMMetric
|
| 164 |
+
name: SSIM
|
| 165 |
+
- _target_: src.metrics.GMSDMetric
|
| 166 |
+
name: GMSD
|
| 167 |
+
- _target_: src.metrics.MSEMetric
|
| 168 |
+
name: MSE
|
| 169 |
+
normalized: false
|
| 170 |
+
- _target_: src.metrics.MSEMetric
|
| 171 |
+
name: NormMSE
|
| 172 |
+
normalized: true
|
| 173 |
+
datasets:
|
| 174 |
+
train:
|
| 175 |
+
_target_: src.datasets.LibrispeechDataset
|
| 176 |
+
max_audio_length: 3
|
| 177 |
+
part: train-clean-100
|
| 178 |
+
roi_kwargs: ${reconstruction.roi_kwargs}
|
| 179 |
+
codec_name: ${codec.codec_name}
|
| 180 |
+
lensless_tag: measurement
|
| 181 |
+
instance_transforms: ${transforms.instance_transforms.train}
|
| 182 |
+
sim_psf_config: ${psf}
|
| 183 |
+
test:
|
| 184 |
+
_target_: src.datasets.LibrispeechDataset
|
| 185 |
+
limit: 1
|
| 186 |
+
max_audio_length: 3
|
| 187 |
+
part: test-clean
|
| 188 |
+
roi_kwargs: ${reconstruction.roi_kwargs}
|
| 189 |
+
codec_name: ${codec.codec_name}
|
| 190 |
+
lensless_tag: measurement
|
| 191 |
+
instance_transforms: ${transforms.instance_transforms.inference}
|
| 192 |
+
sim_psf_config: ${psf}
|
| 193 |
+
music_test:
|
| 194 |
+
_target_: src.datasets.SongDescriberDataset
|
| 195 |
+
limit: 1
|
| 196 |
+
part: test
|
| 197 |
+
roi_kwargs: ${reconstruction.roi_kwargs}
|
| 198 |
+
codec_name: ${codec.codec_name}
|
| 199 |
+
lensless_tag: measurement
|
| 200 |
+
instance_transforms: ${transforms.instance_transforms.inference}
|
| 201 |
+
sim_psf_config: ${psf}
|
| 202 |
+
dataloader:
|
| 203 |
+
train:
|
| 204 |
+
_target_: torch.utils.data.DataLoader
|
| 205 |
+
batch_size: 1
|
| 206 |
+
num_workers: 2
|
| 207 |
+
pin_memory: true
|
| 208 |
+
inference:
|
| 209 |
+
_target_: torch.utils.data.DataLoader
|
| 210 |
+
batch_size: 1
|
| 211 |
+
num_workers: 2
|
| 212 |
+
pin_memory: true
|
| 213 |
+
transforms:
|
| 214 |
+
instance_transforms:
|
| 215 |
+
train:
|
| 216 |
+
all:
|
| 217 |
+
_target_: torchvision.transforms.v2.Compose
|
| 218 |
+
transforms:
|
| 219 |
+
- _target_: src.transforms.PadCrop
|
| 220 |
+
length: 4
|
| 221 |
+
pad_format: replicated
|
| 222 |
+
random_crop: true
|
| 223 |
+
inference: null
|
| 224 |
+
batch_transforms:
|
| 225 |
+
train: null
|
| 226 |
+
inference: null
|
| 227 |
+
codec:
|
| 228 |
+
_target_: src.transforms.CodecEncoderDecoder
|
| 229 |
+
codec_cls: ${resolve_class:dac.DAC}
|
| 230 |
+
codec_weights_path: data/dac_exps/${codec.codec_name}/latest/dac/weights.pth
|
| 231 |
+
codec_add_root_path: true
|
| 232 |
+
codec_kwargs: null
|
| 233 |
+
codec_name: 32x32_120_16khz_original
|
| 234 |
+
eval_mode: true
|
| 235 |
+
freeze_weights: true
|
| 236 |
+
reconstruction:
|
| 237 |
+
roi_kwargs:
|
| 238 |
+
top_left:
|
| 239 |
+
- 65
|
| 240 |
+
- 118
|
| 241 |
+
height: 256
|
| 242 |
+
width: 256
|
| 243 |
+
resize_coef: 8
|
| 244 |
+
normalize_lensless: true
|
| 245 |
+
corners_list: null
|
| 246 |
+
psf:
|
| 247 |
+
slm: adafruit
|
| 248 |
+
sensor: rpi_hq
|
| 249 |
+
downsample: 8
|
| 250 |
+
rotate: -0.8
|
| 251 |
+
vertical_shift: -20
|
| 252 |
+
horizontal_shift: -20
|
| 253 |
+
flipud: true
|
| 254 |
+
use_waveprop: true
|
| 255 |
+
deadspace: true
|
| 256 |
+
scene2mask: 0.3
|
| 257 |
+
mask2sensor: 0.004
|
| 258 |
+
grayscale: true
|
| 259 |
+
lr_scheduler:
|
| 260 |
+
_target_: torch.optim.lr_scheduler.ConstantLR
|
| 261 |
+
factor: 1
|
| 262 |
+
optimizer:
|
| 263 |
+
_target_: torch.optim.Adam
|
| 264 |
+
lr: 0.0001
|
| 265 |
+
loss_function:
|
| 266 |
+
_target_: src.loss.ReconstructionLoss
|
| 267 |
+
codec_mse_coef: 1
|
| 268 |
+
codec_ssim_coef: 0
|
| 269 |
+
codec_gmsd_coef: 0
|
| 270 |
+
raw_codec_ssim_coef: 0
|
| 271 |
+
raw_codec_l1_coef: 0
|
| 272 |
+
audio_l1_coef: 0
|
| 273 |
+
audio_sisdr_coef: 0
|
| 274 |
+
audio_stft_coef: 0
|
| 275 |
+
audio_mel_coef: 0
|
| 276 |
+
audio_stft_config:
|
| 277 |
+
window_lengths:
|
| 278 |
+
- 2048
|
| 279 |
+
- 512
|
| 280 |
+
audio_mel_config:
|
| 281 |
+
n_mels:
|
| 282 |
+
- 5
|
| 283 |
+
- 10
|
| 284 |
+
- 20
|
| 285 |
+
- 40
|
| 286 |
+
window_lengths:
|
| 287 |
+
- 32
|
| 288 |
+
- 64
|
| 289 |
+
- 128
|
| 290 |
+
- 256
|
| 291 |
+
mel_fmin:
|
| 292 |
+
- 0
|
| 293 |
+
- 0
|
| 294 |
+
- 0
|
| 295 |
+
- 0
|
| 296 |
+
mel_fmax:
|
| 297 |
+
- null
|
| 298 |
+
- null
|
| 299 |
+
- null
|
| 300 |
+
- null
|
| 301 |
+
pow: 1.0
|
| 302 |
+
clamp_eps: 1.0e-05
|
| 303 |
+
mag_weight: 0.0
|
| 304 |
+
resize_coef: ${reconstruction.resize_coef}
|
| 305 |
+
ssim_kernel: 7
|
| 306 |
+
ssim_sigma: 0.5
|
| 307 |
+
raw_ssim_kernel: 11
|
| 308 |
+
trainer:
|
| 309 |
+
log_step: 50
|
| 310 |
+
n_epochs: 100
|
| 311 |
+
epoch_len: 500
|
| 312 |
+
device_tensors:
|
| 313 |
+
- lensless_codec_video
|
| 314 |
+
- lensed_codec_video
|
| 315 |
+
- lensless_psf
|
| 316 |
+
- audio
|
| 317 |
+
resume_from: null
|
| 318 |
+
device: auto
|
| 319 |
+
override: true
|
| 320 |
+
monitor: max test_PSNR
|
| 321 |
+
save_period: 5
|
| 322 |
+
early_stop: ${trainer.n_epochs}
|
| 323 |
+
save_dir: saved
|
| 324 |
+
seed: 1
|
32x32_librispeech_mse_ssim_PSF_Unet4M_U5_Unet4M/checkpoint-epoch100.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:31f9b2638fbac8a9a872573ee632c2dd2afa875958ad68a072898967b3543d08
|
| 3 |
+
size 97418426
|
32x32_librispeech_mse_ssim_PSF_Unet4M_U5_Unet4M/config.yaml
ADDED
|
@@ -0,0 +1,324 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
model:
|
| 2 |
+
_target_: src.model.LenslessWrapper
|
| 3 |
+
use_loader: false
|
| 4 |
+
loader_kwargs: null
|
| 5 |
+
use_batch_video_version: false
|
| 6 |
+
freeze_weights: false
|
| 7 |
+
psf_path: data/digicam_psf/SIM_psf.png
|
| 8 |
+
psf_loader_kwargs:
|
| 9 |
+
downsample: 8
|
| 10 |
+
return_bg: false
|
| 11 |
+
grayscale_psf: true
|
| 12 |
+
recon_name: UnrolledADMM
|
| 13 |
+
recon_kwargs:
|
| 14 |
+
n_iter: 5
|
| 15 |
+
mu1: 0.0001
|
| 16 |
+
mu2: 0.0001
|
| 17 |
+
mu3: 0.0001
|
| 18 |
+
tau: 0.0002
|
| 19 |
+
pre_process:
|
| 20 |
+
_target_: lensless.recon.drunet.network_unet.UNetRes
|
| 21 |
+
in_nc: 2
|
| 22 |
+
out_nc: 1
|
| 23 |
+
nc:
|
| 24 |
+
- 32
|
| 25 |
+
- 64
|
| 26 |
+
- 112
|
| 27 |
+
- 128
|
| 28 |
+
nb: 4
|
| 29 |
+
act_mode: R
|
| 30 |
+
downsample_mode: strideconv
|
| 31 |
+
upsample_mode: convtranspose
|
| 32 |
+
post_process:
|
| 33 |
+
_target_: lensless.recon.drunet.network_unet.UNetRes
|
| 34 |
+
in_nc: 2
|
| 35 |
+
out_nc: 1
|
| 36 |
+
nc:
|
| 37 |
+
- 32
|
| 38 |
+
- 64
|
| 39 |
+
- 116
|
| 40 |
+
- 128
|
| 41 |
+
nb: 4
|
| 42 |
+
act_mode: R
|
| 43 |
+
downsample_mode: strideconv
|
| 44 |
+
upsample_mode: convtranspose
|
| 45 |
+
psf_network:
|
| 46 |
+
_target_: lensless.recon.drunet.network_unet.UNetRes
|
| 47 |
+
in_nc: 2
|
| 48 |
+
out_nc: 1
|
| 49 |
+
nc:
|
| 50 |
+
- 4
|
| 51 |
+
- 8
|
| 52 |
+
- 16
|
| 53 |
+
- 32
|
| 54 |
+
nb: 4
|
| 55 |
+
act_mode: R
|
| 56 |
+
downsample_mode: strideconv
|
| 57 |
+
upsample_mode: convtranspose
|
| 58 |
+
psf_residual: false
|
| 59 |
+
skip_unrolled: false
|
| 60 |
+
return_intermediate: false
|
| 61 |
+
writer:
|
| 62 |
+
_target_: src.logger.WandBWriter
|
| 63 |
+
project_name: lenslessmic
|
| 64 |
+
entity: null
|
| 65 |
+
run_name: 32x32_librispeech_mse_ssim_PSF_Unet4M_U5_Unet4M
|
| 66 |
+
mode: online
|
| 67 |
+
loss_names:
|
| 68 |
+
- loss
|
| 69 |
+
- codec_mse_loss
|
| 70 |
+
- codec_ssim_loss
|
| 71 |
+
- raw_codec_ssim_loss
|
| 72 |
+
- raw_codec_l1_loss
|
| 73 |
+
- audio_l1_loss
|
| 74 |
+
- audio_sisdr_loss
|
| 75 |
+
- audio_stft_loss
|
| 76 |
+
- audio_mel_loss
|
| 77 |
+
log_checkpoints: false
|
| 78 |
+
id_length: 8
|
| 79 |
+
names:
|
| 80 |
+
- input-1:frame
|
| 81 |
+
- input-2:frame
|
| 82 |
+
- input-3:frame
|
| 83 |
+
- input-4:frame
|
| 84 |
+
figsize:
|
| 85 |
+
- 15
|
| 86 |
+
- 15
|
| 87 |
+
sample_rate: 16000
|
| 88 |
+
run_id: y1ya028x
|
| 89 |
+
metrics:
|
| 90 |
+
device: auto
|
| 91 |
+
train:
|
| 92 |
+
- _target_: src.metrics.SISDRMetric
|
| 93 |
+
name: SISDR
|
| 94 |
+
- _target_: src.metrics.PSNRMetric
|
| 95 |
+
name: PSNR
|
| 96 |
+
- _target_: src.metrics.QuantizationMatchMetric
|
| 97 |
+
name: QuantizationMatch-all
|
| 98 |
+
codebook_index: all
|
| 99 |
+
inference:
|
| 100 |
+
- _target_: src.metrics.SISDRMetric
|
| 101 |
+
name: SISDR
|
| 102 |
+
- _target_: src.metrics.STOIMetric
|
| 103 |
+
name: STOI
|
| 104 |
+
- _target_: src.metrics.WERMetric
|
| 105 |
+
name: WER
|
| 106 |
+
- _target_: src.metrics.PESQMetric
|
| 107 |
+
name: PESQ
|
| 108 |
+
- _target_: src.metrics.MelMetric
|
| 109 |
+
name: Mel
|
| 110 |
+
audio_mel_config:
|
| 111 |
+
n_mels:
|
| 112 |
+
- 5
|
| 113 |
+
- 10
|
| 114 |
+
- 20
|
| 115 |
+
- 40
|
| 116 |
+
- 80
|
| 117 |
+
- 160
|
| 118 |
+
- 320
|
| 119 |
+
window_lengths:
|
| 120 |
+
- 32
|
| 121 |
+
- 64
|
| 122 |
+
- 128
|
| 123 |
+
- 256
|
| 124 |
+
- 512
|
| 125 |
+
- 1024
|
| 126 |
+
- 2048
|
| 127 |
+
mel_fmin:
|
| 128 |
+
- 0
|
| 129 |
+
- 0
|
| 130 |
+
- 0
|
| 131 |
+
- 0
|
| 132 |
+
- 0
|
| 133 |
+
- 0
|
| 134 |
+
- 0
|
| 135 |
+
mel_fmax:
|
| 136 |
+
- null
|
| 137 |
+
- null
|
| 138 |
+
- null
|
| 139 |
+
- null
|
| 140 |
+
- null
|
| 141 |
+
- null
|
| 142 |
+
- null
|
| 143 |
+
pow: 1.0
|
| 144 |
+
clamp_eps: 1.0e-05
|
| 145 |
+
mag_weight: 0.0
|
| 146 |
+
- _target_: src.metrics.STFTMetric
|
| 147 |
+
name: STFT
|
| 148 |
+
audio_stft_config:
|
| 149 |
+
window_lengths:
|
| 150 |
+
- 2048
|
| 151 |
+
- 512
|
| 152 |
+
- _target_: src.metrics.QuantizationMatchMetric
|
| 153 |
+
name: QuantizationMatch-all
|
| 154 |
+
codebook_index: all
|
| 155 |
+
- _target_: src.metrics.QuantizationMatchMetric
|
| 156 |
+
name: QuantizationMatch-1
|
| 157 |
+
codebook_index: 1
|
| 158 |
+
- _target_: src.metrics.QuantizationMatchMetric
|
| 159 |
+
name: QuantizationMatch-2
|
| 160 |
+
codebook_index: 2
|
| 161 |
+
- _target_: src.metrics.PSNRMetric
|
| 162 |
+
name: PSNR
|
| 163 |
+
- _target_: src.metrics.SSIMMetric
|
| 164 |
+
name: SSIM
|
| 165 |
+
- _target_: src.metrics.GMSDMetric
|
| 166 |
+
name: GMSD
|
| 167 |
+
- _target_: src.metrics.MSEMetric
|
| 168 |
+
name: MSE
|
| 169 |
+
normalized: false
|
| 170 |
+
- _target_: src.metrics.MSEMetric
|
| 171 |
+
name: NormMSE
|
| 172 |
+
normalized: true
|
| 173 |
+
datasets:
|
| 174 |
+
train:
|
| 175 |
+
_target_: src.datasets.LibrispeechDataset
|
| 176 |
+
max_audio_length: 3
|
| 177 |
+
part: train-clean-100
|
| 178 |
+
roi_kwargs: ${reconstruction.roi_kwargs}
|
| 179 |
+
codec_name: ${codec.codec_name}
|
| 180 |
+
lensless_tag: measurement
|
| 181 |
+
instance_transforms: ${transforms.instance_transforms.train}
|
| 182 |
+
sim_psf_config: ${psf}
|
| 183 |
+
test:
|
| 184 |
+
_target_: src.datasets.LibrispeechDataset
|
| 185 |
+
limit: 1
|
| 186 |
+
max_audio_length: 3
|
| 187 |
+
part: test-clean
|
| 188 |
+
roi_kwargs: ${reconstruction.roi_kwargs}
|
| 189 |
+
codec_name: ${codec.codec_name}
|
| 190 |
+
lensless_tag: measurement
|
| 191 |
+
instance_transforms: ${transforms.instance_transforms.inference}
|
| 192 |
+
sim_psf_config: ${psf}
|
| 193 |
+
music_test:
|
| 194 |
+
_target_: src.datasets.SongDescriberDataset
|
| 195 |
+
limit: 1
|
| 196 |
+
part: test
|
| 197 |
+
roi_kwargs: ${reconstruction.roi_kwargs}
|
| 198 |
+
codec_name: ${codec.codec_name}
|
| 199 |
+
lensless_tag: measurement
|
| 200 |
+
instance_transforms: ${transforms.instance_transforms.inference}
|
| 201 |
+
sim_psf_config: ${psf}
|
| 202 |
+
dataloader:
|
| 203 |
+
train:
|
| 204 |
+
_target_: torch.utils.data.DataLoader
|
| 205 |
+
batch_size: 1
|
| 206 |
+
num_workers: 2
|
| 207 |
+
pin_memory: true
|
| 208 |
+
inference:
|
| 209 |
+
_target_: torch.utils.data.DataLoader
|
| 210 |
+
batch_size: 1
|
| 211 |
+
num_workers: 2
|
| 212 |
+
pin_memory: true
|
| 213 |
+
transforms:
|
| 214 |
+
instance_transforms:
|
| 215 |
+
train:
|
| 216 |
+
all:
|
| 217 |
+
_target_: torchvision.transforms.v2.Compose
|
| 218 |
+
transforms:
|
| 219 |
+
- _target_: src.transforms.PadCrop
|
| 220 |
+
length: 4
|
| 221 |
+
pad_format: replicated
|
| 222 |
+
random_crop: true
|
| 223 |
+
inference: null
|
| 224 |
+
batch_transforms:
|
| 225 |
+
train: null
|
| 226 |
+
inference: null
|
| 227 |
+
codec:
|
| 228 |
+
_target_: src.transforms.CodecEncoderDecoder
|
| 229 |
+
codec_cls: ${resolve_class:dac.DAC}
|
| 230 |
+
codec_weights_path: data/dac_exps/${codec.codec_name}/latest/dac/weights.pth
|
| 231 |
+
codec_add_root_path: true
|
| 232 |
+
codec_kwargs: null
|
| 233 |
+
codec_name: 32x32_120_16khz_original
|
| 234 |
+
eval_mode: true
|
| 235 |
+
freeze_weights: true
|
| 236 |
+
reconstruction:
|
| 237 |
+
roi_kwargs:
|
| 238 |
+
top_left:
|
| 239 |
+
- 65
|
| 240 |
+
- 118
|
| 241 |
+
height: 256
|
| 242 |
+
width: 256
|
| 243 |
+
resize_coef: 8
|
| 244 |
+
normalize_lensless: true
|
| 245 |
+
corners_list: null
|
| 246 |
+
psf:
|
| 247 |
+
slm: adafruit
|
| 248 |
+
sensor: rpi_hq
|
| 249 |
+
downsample: 8
|
| 250 |
+
rotate: -0.8
|
| 251 |
+
vertical_shift: -20
|
| 252 |
+
horizontal_shift: -20
|
| 253 |
+
flipud: true
|
| 254 |
+
use_waveprop: true
|
| 255 |
+
deadspace: true
|
| 256 |
+
scene2mask: 0.3
|
| 257 |
+
mask2sensor: 0.004
|
| 258 |
+
grayscale: true
|
| 259 |
+
lr_scheduler:
|
| 260 |
+
_target_: torch.optim.lr_scheduler.ConstantLR
|
| 261 |
+
factor: 1
|
| 262 |
+
optimizer:
|
| 263 |
+
_target_: torch.optim.Adam
|
| 264 |
+
lr: 0.0001
|
| 265 |
+
loss_function:
|
| 266 |
+
_target_: src.loss.ReconstructionLoss
|
| 267 |
+
codec_mse_coef: 1
|
| 268 |
+
codec_ssim_coef: 1
|
| 269 |
+
codec_gmsd_coef: 0
|
| 270 |
+
raw_codec_ssim_coef: 0
|
| 271 |
+
raw_codec_l1_coef: 0
|
| 272 |
+
audio_l1_coef: 0
|
| 273 |
+
audio_sisdr_coef: 0
|
| 274 |
+
audio_stft_coef: 0
|
| 275 |
+
audio_mel_coef: 0
|
| 276 |
+
audio_stft_config:
|
| 277 |
+
window_lengths:
|
| 278 |
+
- 2048
|
| 279 |
+
- 512
|
| 280 |
+
audio_mel_config:
|
| 281 |
+
n_mels:
|
| 282 |
+
- 5
|
| 283 |
+
- 10
|
| 284 |
+
- 20
|
| 285 |
+
- 40
|
| 286 |
+
window_lengths:
|
| 287 |
+
- 32
|
| 288 |
+
- 64
|
| 289 |
+
- 128
|
| 290 |
+
- 256
|
| 291 |
+
mel_fmin:
|
| 292 |
+
- 0
|
| 293 |
+
- 0
|
| 294 |
+
- 0
|
| 295 |
+
- 0
|
| 296 |
+
mel_fmax:
|
| 297 |
+
- null
|
| 298 |
+
- null
|
| 299 |
+
- null
|
| 300 |
+
- null
|
| 301 |
+
pow: 1.0
|
| 302 |
+
clamp_eps: 1.0e-05
|
| 303 |
+
mag_weight: 0.0
|
| 304 |
+
resize_coef: ${reconstruction.resize_coef}
|
| 305 |
+
ssim_kernel: 7
|
| 306 |
+
ssim_sigma: 0.5
|
| 307 |
+
raw_ssim_kernel: 11
|
| 308 |
+
trainer:
|
| 309 |
+
log_step: 50
|
| 310 |
+
n_epochs: 100
|
| 311 |
+
epoch_len: 500
|
| 312 |
+
device_tensors:
|
| 313 |
+
- lensless_codec_video
|
| 314 |
+
- lensed_codec_video
|
| 315 |
+
- lensless_psf
|
| 316 |
+
- audio
|
| 317 |
+
resume_from: null
|
| 318 |
+
device: auto
|
| 319 |
+
override: true
|
| 320 |
+
monitor: max test_PSNR
|
| 321 |
+
save_period: 5
|
| 322 |
+
early_stop: ${trainer.n_epochs}
|
| 323 |
+
save_dir: saved
|
| 324 |
+
seed: 1
|
32x32_librispeech_mse_ssim_raw_ssim_PSF_Unet4M_U5_Unet4M/checkpoint-epoch100.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f5c207fdf5ff4cc38d37d1a3a2117beba802b5f2f574920ac5808277688b281c
|
| 3 |
+
size 97416762
|
32x32_librispeech_mse_ssim_raw_ssim_PSF_Unet4M_U5_Unet4M/config.yaml
ADDED
|
@@ -0,0 +1,315 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
model:
|
| 2 |
+
_target_: src.model.LenslessWrapper
|
| 3 |
+
use_loader: false
|
| 4 |
+
loader_kwargs: null
|
| 5 |
+
use_batch_video_version: false
|
| 6 |
+
freeze_weights: false
|
| 7 |
+
psf_path: data/digicam_psf/SIM_psf.png
|
| 8 |
+
psf_loader_kwargs:
|
| 9 |
+
downsample: 8
|
| 10 |
+
return_bg: false
|
| 11 |
+
grayscale_psf: true
|
| 12 |
+
recon_name: UnrolledADMM
|
| 13 |
+
recon_kwargs:
|
| 14 |
+
n_iter: 5
|
| 15 |
+
mu1: 0.0001
|
| 16 |
+
mu2: 0.0001
|
| 17 |
+
mu3: 0.0001
|
| 18 |
+
tau: 0.0002
|
| 19 |
+
pre_process:
|
| 20 |
+
_target_: lensless.recon.drunet.network_unet.UNetRes
|
| 21 |
+
in_nc: 2
|
| 22 |
+
out_nc: 1
|
| 23 |
+
nc:
|
| 24 |
+
- 32
|
| 25 |
+
- 64
|
| 26 |
+
- 112
|
| 27 |
+
- 128
|
| 28 |
+
nb: 4
|
| 29 |
+
act_mode: R
|
| 30 |
+
downsample_mode: strideconv
|
| 31 |
+
upsample_mode: convtranspose
|
| 32 |
+
post_process:
|
| 33 |
+
_target_: lensless.recon.drunet.network_unet.UNetRes
|
| 34 |
+
in_nc: 2
|
| 35 |
+
out_nc: 1
|
| 36 |
+
nc:
|
| 37 |
+
- 32
|
| 38 |
+
- 64
|
| 39 |
+
- 116
|
| 40 |
+
- 128
|
| 41 |
+
nb: 4
|
| 42 |
+
act_mode: R
|
| 43 |
+
downsample_mode: strideconv
|
| 44 |
+
upsample_mode: convtranspose
|
| 45 |
+
psf_network:
|
| 46 |
+
_target_: lensless.recon.drunet.network_unet.UNetRes
|
| 47 |
+
in_nc: 2
|
| 48 |
+
out_nc: 1
|
| 49 |
+
nc:
|
| 50 |
+
- 4
|
| 51 |
+
- 8
|
| 52 |
+
- 16
|
| 53 |
+
- 32
|
| 54 |
+
nb: 4
|
| 55 |
+
act_mode: R
|
| 56 |
+
downsample_mode: strideconv
|
| 57 |
+
upsample_mode: convtranspose
|
| 58 |
+
psf_residual: false
|
| 59 |
+
skip_unrolled: false
|
| 60 |
+
return_intermediate: false
|
| 61 |
+
writer:
|
| 62 |
+
_target_: src.logger.WandBWriter
|
| 63 |
+
project_name: lenslessmic
|
| 64 |
+
entity: null
|
| 65 |
+
run_name: 32x32_librispeech_mse_ssim_raw_ssim_PSF_Unet4M_U5_Unet4M
|
| 66 |
+
mode: online
|
| 67 |
+
loss_names:
|
| 68 |
+
- loss
|
| 69 |
+
- codec_mse_loss
|
| 70 |
+
- codec_ssim_loss
|
| 71 |
+
- raw_codec_ssim_loss
|
| 72 |
+
- raw_codec_l1_loss
|
| 73 |
+
- audio_l1_loss
|
| 74 |
+
- audio_sisdr_loss
|
| 75 |
+
- audio_stft_loss
|
| 76 |
+
- audio_mel_loss
|
| 77 |
+
log_checkpoints: false
|
| 78 |
+
id_length: 8
|
| 79 |
+
names:
|
| 80 |
+
- input-1:frame
|
| 81 |
+
- input-2:frame
|
| 82 |
+
- input-3:frame
|
| 83 |
+
- input-4:frame
|
| 84 |
+
figsize:
|
| 85 |
+
- 15
|
| 86 |
+
- 15
|
| 87 |
+
sample_rate: 16000
|
| 88 |
+
run_id: 3j7yioen
|
| 89 |
+
metrics:
|
| 90 |
+
device: auto
|
| 91 |
+
train:
|
| 92 |
+
- _target_: src.metrics.SISDRMetric
|
| 93 |
+
name: SISDR
|
| 94 |
+
- _target_: src.metrics.PSNRMetric
|
| 95 |
+
name: PSNR
|
| 96 |
+
- _target_: src.metrics.QuantizationMatchMetric
|
| 97 |
+
name: QuantizationMatch-all
|
| 98 |
+
codebook_index: all
|
| 99 |
+
inference:
|
| 100 |
+
- _target_: src.metrics.SISDRMetric
|
| 101 |
+
name: SISDR
|
| 102 |
+
- _target_: src.metrics.STOIMetric
|
| 103 |
+
name: STOI
|
| 104 |
+
- _target_: src.metrics.WERMetric
|
| 105 |
+
name: WER
|
| 106 |
+
- _target_: src.metrics.PESQMetric
|
| 107 |
+
name: PESQ
|
| 108 |
+
- _target_: src.metrics.MelMetric
|
| 109 |
+
name: Mel
|
| 110 |
+
audio_mel_config:
|
| 111 |
+
n_mels:
|
| 112 |
+
- 5
|
| 113 |
+
- 10
|
| 114 |
+
- 20
|
| 115 |
+
- 40
|
| 116 |
+
- 80
|
| 117 |
+
- 160
|
| 118 |
+
- 320
|
| 119 |
+
window_lengths:
|
| 120 |
+
- 32
|
| 121 |
+
- 64
|
| 122 |
+
- 128
|
| 123 |
+
- 256
|
| 124 |
+
- 512
|
| 125 |
+
- 1024
|
| 126 |
+
- 2048
|
| 127 |
+
mel_fmin:
|
| 128 |
+
- 0
|
| 129 |
+
- 0
|
| 130 |
+
- 0
|
| 131 |
+
- 0
|
| 132 |
+
- 0
|
| 133 |
+
- 0
|
| 134 |
+
- 0
|
| 135 |
+
mel_fmax:
|
| 136 |
+
- null
|
| 137 |
+
- null
|
| 138 |
+
- null
|
| 139 |
+
- null
|
| 140 |
+
- null
|
| 141 |
+
- null
|
| 142 |
+
- null
|
| 143 |
+
pow: 1.0
|
| 144 |
+
clamp_eps: 1.0e-05
|
| 145 |
+
mag_weight: 0.0
|
| 146 |
+
- _target_: src.metrics.STFTMetric
|
| 147 |
+
name: STFT
|
| 148 |
+
audio_stft_config:
|
| 149 |
+
window_lengths:
|
| 150 |
+
- 2048
|
| 151 |
+
- 512
|
| 152 |
+
- _target_: src.metrics.QuantizationMatchMetric
|
| 153 |
+
name: QuantizationMatch-all
|
| 154 |
+
codebook_index: all
|
| 155 |
+
- _target_: src.metrics.QuantizationMatchMetric
|
| 156 |
+
name: QuantizationMatch-1
|
| 157 |
+
codebook_index: 1
|
| 158 |
+
- _target_: src.metrics.QuantizationMatchMetric
|
| 159 |
+
name: QuantizationMatch-2
|
| 160 |
+
codebook_index: 2
|
| 161 |
+
- _target_: src.metrics.PSNRMetric
|
| 162 |
+
name: PSNR
|
| 163 |
+
- _target_: src.metrics.SSIMMetric
|
| 164 |
+
name: SSIM
|
| 165 |
+
- _target_: src.metrics.GMSDMetric
|
| 166 |
+
name: GMSD
|
| 167 |
+
- _target_: src.metrics.MSEMetric
|
| 168 |
+
name: MSE
|
| 169 |
+
normalized: false
|
| 170 |
+
- _target_: src.metrics.MSEMetric
|
| 171 |
+
name: NormMSE
|
| 172 |
+
normalized: true
|
| 173 |
+
datasets:
|
| 174 |
+
train:
|
| 175 |
+
_target_: src.datasets.LibrispeechDataset
|
| 176 |
+
max_audio_length: 3
|
| 177 |
+
part: train-clean-100
|
| 178 |
+
roi_kwargs: ${reconstruction.roi_kwargs}
|
| 179 |
+
codec_name: ${codec.codec_name}
|
| 180 |
+
lensless_tag: measurement
|
| 181 |
+
instance_transforms: ${transforms.instance_transforms.train}
|
| 182 |
+
sim_psf_config: ${psf}
|
| 183 |
+
test:
|
| 184 |
+
_target_: src.datasets.LibrispeechDataset
|
| 185 |
+
limit: 1
|
| 186 |
+
max_audio_length: 3
|
| 187 |
+
part: test-clean
|
| 188 |
+
roi_kwargs: ${reconstruction.roi_kwargs}
|
| 189 |
+
codec_name: ${codec.codec_name}
|
| 190 |
+
lensless_tag: measurement
|
| 191 |
+
instance_transforms: ${transforms.instance_transforms.inference}
|
| 192 |
+
sim_psf_config: ${psf}
|
| 193 |
+
dataloader:
|
| 194 |
+
train:
|
| 195 |
+
_target_: torch.utils.data.DataLoader
|
| 196 |
+
batch_size: 1
|
| 197 |
+
num_workers: 2
|
| 198 |
+
pin_memory: true
|
| 199 |
+
inference:
|
| 200 |
+
_target_: torch.utils.data.DataLoader
|
| 201 |
+
batch_size: 1
|
| 202 |
+
num_workers: 2
|
| 203 |
+
pin_memory: true
|
| 204 |
+
transforms:
|
| 205 |
+
instance_transforms:
|
| 206 |
+
train:
|
| 207 |
+
all:
|
| 208 |
+
_target_: torchvision.transforms.v2.Compose
|
| 209 |
+
transforms:
|
| 210 |
+
- _target_: src.transforms.PadCrop
|
| 211 |
+
length: 4
|
| 212 |
+
pad_format: replicated
|
| 213 |
+
random_crop: true
|
| 214 |
+
inference: null
|
| 215 |
+
batch_transforms:
|
| 216 |
+
train: null
|
| 217 |
+
inference: null
|
| 218 |
+
codec:
|
| 219 |
+
_target_: src.transforms.CodecEncoderDecoder
|
| 220 |
+
codec_cls: ${resolve_class:dac.DAC}
|
| 221 |
+
codec_weights_path: data/dac_exps/${codec.codec_name}/latest/dac/weights.pth
|
| 222 |
+
codec_add_root_path: true
|
| 223 |
+
codec_kwargs: null
|
| 224 |
+
codec_name: 32x32_120_16khz_original
|
| 225 |
+
eval_mode: true
|
| 226 |
+
freeze_weights: true
|
| 227 |
+
reconstruction:
|
| 228 |
+
roi_kwargs:
|
| 229 |
+
top_left:
|
| 230 |
+
- 65
|
| 231 |
+
- 118
|
| 232 |
+
height: 256
|
| 233 |
+
width: 256
|
| 234 |
+
resize_coef: 8
|
| 235 |
+
normalize_lensless: true
|
| 236 |
+
corners_list: null
|
| 237 |
+
psf:
|
| 238 |
+
slm: adafruit
|
| 239 |
+
sensor: rpi_hq
|
| 240 |
+
downsample: 8
|
| 241 |
+
rotate: -0.8
|
| 242 |
+
vertical_shift: -20
|
| 243 |
+
horizontal_shift: -20
|
| 244 |
+
flipud: true
|
| 245 |
+
use_waveprop: true
|
| 246 |
+
deadspace: true
|
| 247 |
+
scene2mask: 0.3
|
| 248 |
+
mask2sensor: 0.004
|
| 249 |
+
grayscale: true
|
| 250 |
+
lr_scheduler:
|
| 251 |
+
_target_: torch.optim.lr_scheduler.ConstantLR
|
| 252 |
+
factor: 1
|
| 253 |
+
optimizer:
|
| 254 |
+
_target_: torch.optim.Adam
|
| 255 |
+
lr: 0.0001
|
| 256 |
+
loss_function:
|
| 257 |
+
_target_: src.loss.ReconstructionLoss
|
| 258 |
+
codec_mse_coef: 1
|
| 259 |
+
codec_ssim_coef: 1
|
| 260 |
+
codec_gmsd_coef: 0
|
| 261 |
+
raw_codec_ssim_coef: 1
|
| 262 |
+
raw_codec_l1_coef: 0
|
| 263 |
+
audio_l1_coef: 0
|
| 264 |
+
audio_sisdr_coef: 0
|
| 265 |
+
audio_stft_coef: 0
|
| 266 |
+
audio_mel_coef: 0
|
| 267 |
+
audio_stft_config:
|
| 268 |
+
window_lengths:
|
| 269 |
+
- 2048
|
| 270 |
+
- 512
|
| 271 |
+
audio_mel_config:
|
| 272 |
+
n_mels:
|
| 273 |
+
- 5
|
| 274 |
+
- 10
|
| 275 |
+
- 20
|
| 276 |
+
- 40
|
| 277 |
+
window_lengths:
|
| 278 |
+
- 32
|
| 279 |
+
- 64
|
| 280 |
+
- 128
|
| 281 |
+
- 256
|
| 282 |
+
mel_fmin:
|
| 283 |
+
- 0
|
| 284 |
+
- 0
|
| 285 |
+
- 0
|
| 286 |
+
- 0
|
| 287 |
+
mel_fmax:
|
| 288 |
+
- null
|
| 289 |
+
- null
|
| 290 |
+
- null
|
| 291 |
+
- null
|
| 292 |
+
pow: 1.0
|
| 293 |
+
clamp_eps: 1.0e-05
|
| 294 |
+
mag_weight: 0.0
|
| 295 |
+
resize_coef: ${reconstruction.resize_coef}
|
| 296 |
+
ssim_kernel: 7
|
| 297 |
+
ssim_sigma: 0.5
|
| 298 |
+
raw_ssim_kernel: 11
|
| 299 |
+
trainer:
|
| 300 |
+
log_step: 50
|
| 301 |
+
n_epochs: 100
|
| 302 |
+
epoch_len: 500
|
| 303 |
+
device_tensors:
|
| 304 |
+
- lensless_codec_video
|
| 305 |
+
- lensed_codec_video
|
| 306 |
+
- lensless_psf
|
| 307 |
+
- audio
|
| 308 |
+
resume_from: null
|
| 309 |
+
device: auto
|
| 310 |
+
override: true
|
| 311 |
+
monitor: max test_PSNR
|
| 312 |
+
save_period: 5
|
| 313 |
+
early_stop: ${trainer.n_epochs}
|
| 314 |
+
save_dir: saved
|
| 315 |
+
seed: 1
|
32x32_librispeech_mse_ssim_raw_ssim_l1_PSF_Unet4M_U5_Unet4M/config.yaml
ADDED
|
@@ -0,0 +1,324 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
model:
|
| 2 |
+
_target_: src.model.LenslessWrapper
|
| 3 |
+
use_loader: false
|
| 4 |
+
loader_kwargs: null
|
| 5 |
+
use_batch_video_version: false
|
| 6 |
+
freeze_weights: false
|
| 7 |
+
psf_path: data/digicam_psf/SIM_psf.png
|
| 8 |
+
psf_loader_kwargs:
|
| 9 |
+
downsample: 8
|
| 10 |
+
return_bg: false
|
| 11 |
+
grayscale_psf: true
|
| 12 |
+
recon_name: UnrolledADMM
|
| 13 |
+
recon_kwargs:
|
| 14 |
+
n_iter: 5
|
| 15 |
+
mu1: 0.0001
|
| 16 |
+
mu2: 0.0001
|
| 17 |
+
mu3: 0.0001
|
| 18 |
+
tau: 0.0002
|
| 19 |
+
pre_process:
|
| 20 |
+
_target_: lensless.recon.drunet.network_unet.UNetRes
|
| 21 |
+
in_nc: 2
|
| 22 |
+
out_nc: 1
|
| 23 |
+
nc:
|
| 24 |
+
- 32
|
| 25 |
+
- 64
|
| 26 |
+
- 112
|
| 27 |
+
- 128
|
| 28 |
+
nb: 4
|
| 29 |
+
act_mode: R
|
| 30 |
+
downsample_mode: strideconv
|
| 31 |
+
upsample_mode: convtranspose
|
| 32 |
+
post_process:
|
| 33 |
+
_target_: lensless.recon.drunet.network_unet.UNetRes
|
| 34 |
+
in_nc: 2
|
| 35 |
+
out_nc: 1
|
| 36 |
+
nc:
|
| 37 |
+
- 32
|
| 38 |
+
- 64
|
| 39 |
+
- 116
|
| 40 |
+
- 128
|
| 41 |
+
nb: 4
|
| 42 |
+
act_mode: R
|
| 43 |
+
downsample_mode: strideconv
|
| 44 |
+
upsample_mode: convtranspose
|
| 45 |
+
psf_network:
|
| 46 |
+
_target_: lensless.recon.drunet.network_unet.UNetRes
|
| 47 |
+
in_nc: 2
|
| 48 |
+
out_nc: 1
|
| 49 |
+
nc:
|
| 50 |
+
- 4
|
| 51 |
+
- 8
|
| 52 |
+
- 16
|
| 53 |
+
- 32
|
| 54 |
+
nb: 4
|
| 55 |
+
act_mode: R
|
| 56 |
+
downsample_mode: strideconv
|
| 57 |
+
upsample_mode: convtranspose
|
| 58 |
+
psf_residual: false
|
| 59 |
+
skip_unrolled: false
|
| 60 |
+
return_intermediate: false
|
| 61 |
+
writer:
|
| 62 |
+
_target_: src.logger.WandBWriter
|
| 63 |
+
project_name: lenslessmic
|
| 64 |
+
entity: null
|
| 65 |
+
run_name: 32x32_librispeech_mse_ssim_raw_ssim_l1_PSF_Unet4M_U5_Unet4M
|
| 66 |
+
mode: online
|
| 67 |
+
loss_names:
|
| 68 |
+
- loss
|
| 69 |
+
- codec_mse_loss
|
| 70 |
+
- codec_ssim_loss
|
| 71 |
+
- raw_codec_ssim_loss
|
| 72 |
+
- raw_codec_l1_loss
|
| 73 |
+
- audio_l1_loss
|
| 74 |
+
- audio_sisdr_loss
|
| 75 |
+
- audio_stft_loss
|
| 76 |
+
- audio_mel_loss
|
| 77 |
+
log_checkpoints: false
|
| 78 |
+
id_length: 8
|
| 79 |
+
names:
|
| 80 |
+
- input-1:frame
|
| 81 |
+
- input-2:frame
|
| 82 |
+
- input-3:frame
|
| 83 |
+
- input-4:frame
|
| 84 |
+
figsize:
|
| 85 |
+
- 15
|
| 86 |
+
- 15
|
| 87 |
+
sample_rate: 16000
|
| 88 |
+
run_id: qsjm2uok
|
| 89 |
+
metrics:
|
| 90 |
+
device: auto
|
| 91 |
+
train:
|
| 92 |
+
- _target_: src.metrics.SISDRMetric
|
| 93 |
+
name: SISDR
|
| 94 |
+
- _target_: src.metrics.PSNRMetric
|
| 95 |
+
name: PSNR
|
| 96 |
+
- _target_: src.metrics.QuantizationMatchMetric
|
| 97 |
+
name: QuantizationMatch-all
|
| 98 |
+
codebook_index: all
|
| 99 |
+
inference:
|
| 100 |
+
- _target_: src.metrics.SISDRMetric
|
| 101 |
+
name: SISDR
|
| 102 |
+
- _target_: src.metrics.STOIMetric
|
| 103 |
+
name: STOI
|
| 104 |
+
- _target_: src.metrics.WERMetric
|
| 105 |
+
name: WER
|
| 106 |
+
- _target_: src.metrics.PESQMetric
|
| 107 |
+
name: PESQ
|
| 108 |
+
- _target_: src.metrics.MelMetric
|
| 109 |
+
name: Mel
|
| 110 |
+
audio_mel_config:
|
| 111 |
+
n_mels:
|
| 112 |
+
- 5
|
| 113 |
+
- 10
|
| 114 |
+
- 20
|
| 115 |
+
- 40
|
| 116 |
+
- 80
|
| 117 |
+
- 160
|
| 118 |
+
- 320
|
| 119 |
+
window_lengths:
|
| 120 |
+
- 32
|
| 121 |
+
- 64
|
| 122 |
+
- 128
|
| 123 |
+
- 256
|
| 124 |
+
- 512
|
| 125 |
+
- 1024
|
| 126 |
+
- 2048
|
| 127 |
+
mel_fmin:
|
| 128 |
+
- 0
|
| 129 |
+
- 0
|
| 130 |
+
- 0
|
| 131 |
+
- 0
|
| 132 |
+
- 0
|
| 133 |
+
- 0
|
| 134 |
+
- 0
|
| 135 |
+
mel_fmax:
|
| 136 |
+
- null
|
| 137 |
+
- null
|
| 138 |
+
- null
|
| 139 |
+
- null
|
| 140 |
+
- null
|
| 141 |
+
- null
|
| 142 |
+
- null
|
| 143 |
+
pow: 1.0
|
| 144 |
+
clamp_eps: 1.0e-05
|
| 145 |
+
mag_weight: 0.0
|
| 146 |
+
- _target_: src.metrics.STFTMetric
|
| 147 |
+
name: STFT
|
| 148 |
+
audio_stft_config:
|
| 149 |
+
window_lengths:
|
| 150 |
+
- 2048
|
| 151 |
+
- 512
|
| 152 |
+
- _target_: src.metrics.QuantizationMatchMetric
|
| 153 |
+
name: QuantizationMatch-all
|
| 154 |
+
codebook_index: all
|
| 155 |
+
- _target_: src.metrics.QuantizationMatchMetric
|
| 156 |
+
name: QuantizationMatch-1
|
| 157 |
+
codebook_index: 1
|
| 158 |
+
- _target_: src.metrics.QuantizationMatchMetric
|
| 159 |
+
name: QuantizationMatch-2
|
| 160 |
+
codebook_index: 2
|
| 161 |
+
- _target_: src.metrics.PSNRMetric
|
| 162 |
+
name: PSNR
|
| 163 |
+
- _target_: src.metrics.SSIMMetric
|
| 164 |
+
name: SSIM
|
| 165 |
+
- _target_: src.metrics.GMSDMetric
|
| 166 |
+
name: GMSD
|
| 167 |
+
- _target_: src.metrics.MSEMetric
|
| 168 |
+
name: MSE
|
| 169 |
+
normalized: false
|
| 170 |
+
- _target_: src.metrics.MSEMetric
|
| 171 |
+
name: NormMSE
|
| 172 |
+
normalized: true
|
| 173 |
+
datasets:
|
| 174 |
+
train:
|
| 175 |
+
_target_: src.datasets.LibrispeechDataset
|
| 176 |
+
max_audio_length: 3
|
| 177 |
+
part: train-clean-100
|
| 178 |
+
roi_kwargs: ${reconstruction.roi_kwargs}
|
| 179 |
+
codec_name: ${codec.codec_name}
|
| 180 |
+
lensless_tag: measurement
|
| 181 |
+
instance_transforms: ${transforms.instance_transforms.train}
|
| 182 |
+
sim_psf_config: ${psf}
|
| 183 |
+
test:
|
| 184 |
+
_target_: src.datasets.LibrispeechDataset
|
| 185 |
+
limit: 1
|
| 186 |
+
max_audio_length: 3
|
| 187 |
+
part: test-clean
|
| 188 |
+
roi_kwargs: ${reconstruction.roi_kwargs}
|
| 189 |
+
codec_name: ${codec.codec_name}
|
| 190 |
+
lensless_tag: measurement
|
| 191 |
+
instance_transforms: ${transforms.instance_transforms.inference}
|
| 192 |
+
sim_psf_config: ${psf}
|
| 193 |
+
music_test:
|
| 194 |
+
_target_: src.datasets.SongDescriberDataset
|
| 195 |
+
limit: 1
|
| 196 |
+
part: test
|
| 197 |
+
roi_kwargs: ${reconstruction.roi_kwargs}
|
| 198 |
+
codec_name: ${codec.codec_name}
|
| 199 |
+
lensless_tag: measurement
|
| 200 |
+
instance_transforms: ${transforms.instance_transforms.inference}
|
| 201 |
+
sim_psf_config: ${psf}
|
| 202 |
+
dataloader:
|
| 203 |
+
train:
|
| 204 |
+
_target_: torch.utils.data.DataLoader
|
| 205 |
+
batch_size: 1
|
| 206 |
+
num_workers: 2
|
| 207 |
+
pin_memory: true
|
| 208 |
+
inference:
|
| 209 |
+
_target_: torch.utils.data.DataLoader
|
| 210 |
+
batch_size: 1
|
| 211 |
+
num_workers: 2
|
| 212 |
+
pin_memory: true
|
| 213 |
+
transforms:
|
| 214 |
+
instance_transforms:
|
| 215 |
+
train:
|
| 216 |
+
all:
|
| 217 |
+
_target_: torchvision.transforms.v2.Compose
|
| 218 |
+
transforms:
|
| 219 |
+
- _target_: src.transforms.PadCrop
|
| 220 |
+
length: 4
|
| 221 |
+
pad_format: replicated
|
| 222 |
+
random_crop: true
|
| 223 |
+
inference: null
|
| 224 |
+
batch_transforms:
|
| 225 |
+
train: null
|
| 226 |
+
inference: null
|
| 227 |
+
codec:
|
| 228 |
+
_target_: src.transforms.CodecEncoderDecoder
|
| 229 |
+
codec_cls: ${resolve_class:dac.DAC}
|
| 230 |
+
codec_weights_path: data/dac_exps/${codec.codec_name}/latest/dac/weights.pth
|
| 231 |
+
codec_add_root_path: true
|
| 232 |
+
codec_kwargs: null
|
| 233 |
+
codec_name: 32x32_120_16khz_original
|
| 234 |
+
eval_mode: true
|
| 235 |
+
freeze_weights: true
|
| 236 |
+
reconstruction:
|
| 237 |
+
roi_kwargs:
|
| 238 |
+
top_left:
|
| 239 |
+
- 65
|
| 240 |
+
- 118
|
| 241 |
+
height: 256
|
| 242 |
+
width: 256
|
| 243 |
+
resize_coef: 8
|
| 244 |
+
normalize_lensless: true
|
| 245 |
+
corners_list: null
|
| 246 |
+
psf:
|
| 247 |
+
slm: adafruit
|
| 248 |
+
sensor: rpi_hq
|
| 249 |
+
downsample: 8
|
| 250 |
+
rotate: -0.8
|
| 251 |
+
vertical_shift: -20
|
| 252 |
+
horizontal_shift: -20
|
| 253 |
+
flipud: true
|
| 254 |
+
use_waveprop: true
|
| 255 |
+
deadspace: true
|
| 256 |
+
scene2mask: 0.3
|
| 257 |
+
mask2sensor: 0.004
|
| 258 |
+
grayscale: true
|
| 259 |
+
lr_scheduler:
|
| 260 |
+
_target_: torch.optim.lr_scheduler.ConstantLR
|
| 261 |
+
factor: 1
|
| 262 |
+
optimizer:
|
| 263 |
+
_target_: torch.optim.Adam
|
| 264 |
+
lr: 0.0001
|
| 265 |
+
loss_function:
|
| 266 |
+
_target_: src.loss.ReconstructionLoss
|
| 267 |
+
codec_mse_coef: 1
|
| 268 |
+
codec_ssim_coef: 1
|
| 269 |
+
codec_gmsd_coef: 0
|
| 270 |
+
raw_codec_ssim_coef: 1
|
| 271 |
+
raw_codec_l1_coef: 0
|
| 272 |
+
audio_l1_coef: 1
|
| 273 |
+
audio_sisdr_coef: 0
|
| 274 |
+
audio_stft_coef: 0
|
| 275 |
+
audio_mel_coef: 0
|
| 276 |
+
audio_stft_config:
|
| 277 |
+
window_lengths:
|
| 278 |
+
- 2048
|
| 279 |
+
- 512
|
| 280 |
+
audio_mel_config:
|
| 281 |
+
n_mels:
|
| 282 |
+
- 5
|
| 283 |
+
- 10
|
| 284 |
+
- 20
|
| 285 |
+
- 40
|
| 286 |
+
window_lengths:
|
| 287 |
+
- 32
|
| 288 |
+
- 64
|
| 289 |
+
- 128
|
| 290 |
+
- 256
|
| 291 |
+
mel_fmin:
|
| 292 |
+
- 0
|
| 293 |
+
- 0
|
| 294 |
+
- 0
|
| 295 |
+
- 0
|
| 296 |
+
mel_fmax:
|
| 297 |
+
- null
|
| 298 |
+
- null
|
| 299 |
+
- null
|
| 300 |
+
- null
|
| 301 |
+
pow: 1.0
|
| 302 |
+
clamp_eps: 1.0e-05
|
| 303 |
+
mag_weight: 0.0
|
| 304 |
+
resize_coef: ${reconstruction.resize_coef}
|
| 305 |
+
ssim_kernel: 7
|
| 306 |
+
ssim_sigma: 0.5
|
| 307 |
+
raw_ssim_kernel: 11
|
| 308 |
+
trainer:
|
| 309 |
+
log_step: 50
|
| 310 |
+
n_epochs: 100
|
| 311 |
+
epoch_len: 500
|
| 312 |
+
device_tensors:
|
| 313 |
+
- lensless_codec_video
|
| 314 |
+
- lensed_codec_video
|
| 315 |
+
- lensless_psf
|
| 316 |
+
- audio
|
| 317 |
+
resume_from: null
|
| 318 |
+
device: auto
|
| 319 |
+
override: true
|
| 320 |
+
monitor: max test_PSNR
|
| 321 |
+
save_period: 5
|
| 322 |
+
early_stop: ${trainer.n_epochs}
|
| 323 |
+
save_dir: saved
|
| 324 |
+
seed: 1
|
32x32_librispeech_mse_ssim_raw_ssim_mel_PSF_Unet4M_U5_Unet4M/checkpoint-epoch100.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:1acf42d38d028f8911caa5385e1136da714f181b62d6c9e88455db8783ff547a
|
| 3 |
+
size 97418426
|
32x32_librispeech_mse_ssim_raw_ssim_mel_PSF_Unet4M_U5_Unet4M/config.yaml
ADDED
|
@@ -0,0 +1,324 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
model:
|
| 2 |
+
_target_: src.model.LenslessWrapper
|
| 3 |
+
use_loader: false
|
| 4 |
+
loader_kwargs: null
|
| 5 |
+
use_batch_video_version: false
|
| 6 |
+
freeze_weights: false
|
| 7 |
+
psf_path: data/digicam_psf/SIM_psf.png
|
| 8 |
+
psf_loader_kwargs:
|
| 9 |
+
downsample: 8
|
| 10 |
+
return_bg: false
|
| 11 |
+
grayscale_psf: true
|
| 12 |
+
recon_name: UnrolledADMM
|
| 13 |
+
recon_kwargs:
|
| 14 |
+
n_iter: 5
|
| 15 |
+
mu1: 0.0001
|
| 16 |
+
mu2: 0.0001
|
| 17 |
+
mu3: 0.0001
|
| 18 |
+
tau: 0.0002
|
| 19 |
+
pre_process:
|
| 20 |
+
_target_: lensless.recon.drunet.network_unet.UNetRes
|
| 21 |
+
in_nc: 2
|
| 22 |
+
out_nc: 1
|
| 23 |
+
nc:
|
| 24 |
+
- 32
|
| 25 |
+
- 64
|
| 26 |
+
- 112
|
| 27 |
+
- 128
|
| 28 |
+
nb: 4
|
| 29 |
+
act_mode: R
|
| 30 |
+
downsample_mode: strideconv
|
| 31 |
+
upsample_mode: convtranspose
|
| 32 |
+
post_process:
|
| 33 |
+
_target_: lensless.recon.drunet.network_unet.UNetRes
|
| 34 |
+
in_nc: 2
|
| 35 |
+
out_nc: 1
|
| 36 |
+
nc:
|
| 37 |
+
- 32
|
| 38 |
+
- 64
|
| 39 |
+
- 116
|
| 40 |
+
- 128
|
| 41 |
+
nb: 4
|
| 42 |
+
act_mode: R
|
| 43 |
+
downsample_mode: strideconv
|
| 44 |
+
upsample_mode: convtranspose
|
| 45 |
+
psf_network:
|
| 46 |
+
_target_: lensless.recon.drunet.network_unet.UNetRes
|
| 47 |
+
in_nc: 2
|
| 48 |
+
out_nc: 1
|
| 49 |
+
nc:
|
| 50 |
+
- 4
|
| 51 |
+
- 8
|
| 52 |
+
- 16
|
| 53 |
+
- 32
|
| 54 |
+
nb: 4
|
| 55 |
+
act_mode: R
|
| 56 |
+
downsample_mode: strideconv
|
| 57 |
+
upsample_mode: convtranspose
|
| 58 |
+
psf_residual: false
|
| 59 |
+
skip_unrolled: false
|
| 60 |
+
return_intermediate: false
|
| 61 |
+
writer:
|
| 62 |
+
_target_: src.logger.WandBWriter
|
| 63 |
+
project_name: lenslessmic
|
| 64 |
+
entity: null
|
| 65 |
+
run_name: 32x32_librispeech_mse_ssim_raw_ssim_mel_PSF_Unet4M_U5_Unet4M
|
| 66 |
+
mode: online
|
| 67 |
+
loss_names:
|
| 68 |
+
- loss
|
| 69 |
+
- codec_mse_loss
|
| 70 |
+
- codec_ssim_loss
|
| 71 |
+
- raw_codec_ssim_loss
|
| 72 |
+
- raw_codec_l1_loss
|
| 73 |
+
- audio_l1_loss
|
| 74 |
+
- audio_sisdr_loss
|
| 75 |
+
- audio_stft_loss
|
| 76 |
+
- audio_mel_loss
|
| 77 |
+
log_checkpoints: false
|
| 78 |
+
id_length: 8
|
| 79 |
+
names:
|
| 80 |
+
- input-1:frame
|
| 81 |
+
- input-2:frame
|
| 82 |
+
- input-3:frame
|
| 83 |
+
- input-4:frame
|
| 84 |
+
figsize:
|
| 85 |
+
- 15
|
| 86 |
+
- 15
|
| 87 |
+
sample_rate: 16000
|
| 88 |
+
run_id: dfjjqazw
|
| 89 |
+
metrics:
|
| 90 |
+
device: auto
|
| 91 |
+
train:
|
| 92 |
+
- _target_: src.metrics.SISDRMetric
|
| 93 |
+
name: SISDR
|
| 94 |
+
- _target_: src.metrics.PSNRMetric
|
| 95 |
+
name: PSNR
|
| 96 |
+
- _target_: src.metrics.QuantizationMatchMetric
|
| 97 |
+
name: QuantizationMatch-all
|
| 98 |
+
codebook_index: all
|
| 99 |
+
inference:
|
| 100 |
+
- _target_: src.metrics.SISDRMetric
|
| 101 |
+
name: SISDR
|
| 102 |
+
- _target_: src.metrics.STOIMetric
|
| 103 |
+
name: STOI
|
| 104 |
+
- _target_: src.metrics.WERMetric
|
| 105 |
+
name: WER
|
| 106 |
+
- _target_: src.metrics.PESQMetric
|
| 107 |
+
name: PESQ
|
| 108 |
+
- _target_: src.metrics.MelMetric
|
| 109 |
+
name: Mel
|
| 110 |
+
audio_mel_config:
|
| 111 |
+
n_mels:
|
| 112 |
+
- 5
|
| 113 |
+
- 10
|
| 114 |
+
- 20
|
| 115 |
+
- 40
|
| 116 |
+
- 80
|
| 117 |
+
- 160
|
| 118 |
+
- 320
|
| 119 |
+
window_lengths:
|
| 120 |
+
- 32
|
| 121 |
+
- 64
|
| 122 |
+
- 128
|
| 123 |
+
- 256
|
| 124 |
+
- 512
|
| 125 |
+
- 1024
|
| 126 |
+
- 2048
|
| 127 |
+
mel_fmin:
|
| 128 |
+
- 0
|
| 129 |
+
- 0
|
| 130 |
+
- 0
|
| 131 |
+
- 0
|
| 132 |
+
- 0
|
| 133 |
+
- 0
|
| 134 |
+
- 0
|
| 135 |
+
mel_fmax:
|
| 136 |
+
- null
|
| 137 |
+
- null
|
| 138 |
+
- null
|
| 139 |
+
- null
|
| 140 |
+
- null
|
| 141 |
+
- null
|
| 142 |
+
- null
|
| 143 |
+
pow: 1.0
|
| 144 |
+
clamp_eps: 1.0e-05
|
| 145 |
+
mag_weight: 0.0
|
| 146 |
+
- _target_: src.metrics.STFTMetric
|
| 147 |
+
name: STFT
|
| 148 |
+
audio_stft_config:
|
| 149 |
+
window_lengths:
|
| 150 |
+
- 2048
|
| 151 |
+
- 512
|
| 152 |
+
- _target_: src.metrics.QuantizationMatchMetric
|
| 153 |
+
name: QuantizationMatch-all
|
| 154 |
+
codebook_index: all
|
| 155 |
+
- _target_: src.metrics.QuantizationMatchMetric
|
| 156 |
+
name: QuantizationMatch-1
|
| 157 |
+
codebook_index: 1
|
| 158 |
+
- _target_: src.metrics.QuantizationMatchMetric
|
| 159 |
+
name: QuantizationMatch-2
|
| 160 |
+
codebook_index: 2
|
| 161 |
+
- _target_: src.metrics.PSNRMetric
|
| 162 |
+
name: PSNR
|
| 163 |
+
- _target_: src.metrics.SSIMMetric
|
| 164 |
+
name: SSIM
|
| 165 |
+
- _target_: src.metrics.GMSDMetric
|
| 166 |
+
name: GMSD
|
| 167 |
+
- _target_: src.metrics.MSEMetric
|
| 168 |
+
name: MSE
|
| 169 |
+
normalized: false
|
| 170 |
+
- _target_: src.metrics.MSEMetric
|
| 171 |
+
name: NormMSE
|
| 172 |
+
normalized: true
|
| 173 |
+
datasets:
|
| 174 |
+
train:
|
| 175 |
+
_target_: src.datasets.LibrispeechDataset
|
| 176 |
+
max_audio_length: 3
|
| 177 |
+
part: train-clean-100
|
| 178 |
+
roi_kwargs: ${reconstruction.roi_kwargs}
|
| 179 |
+
codec_name: ${codec.codec_name}
|
| 180 |
+
lensless_tag: measurement
|
| 181 |
+
instance_transforms: ${transforms.instance_transforms.train}
|
| 182 |
+
sim_psf_config: ${psf}
|
| 183 |
+
test:
|
| 184 |
+
_target_: src.datasets.LibrispeechDataset
|
| 185 |
+
limit: 1
|
| 186 |
+
max_audio_length: 3
|
| 187 |
+
part: test-clean
|
| 188 |
+
roi_kwargs: ${reconstruction.roi_kwargs}
|
| 189 |
+
codec_name: ${codec.codec_name}
|
| 190 |
+
lensless_tag: measurement
|
| 191 |
+
instance_transforms: ${transforms.instance_transforms.inference}
|
| 192 |
+
sim_psf_config: ${psf}
|
| 193 |
+
music_test:
|
| 194 |
+
_target_: src.datasets.SongDescriberDataset
|
| 195 |
+
limit: 1
|
| 196 |
+
part: test
|
| 197 |
+
roi_kwargs: ${reconstruction.roi_kwargs}
|
| 198 |
+
codec_name: ${codec.codec_name}
|
| 199 |
+
lensless_tag: measurement
|
| 200 |
+
instance_transforms: ${transforms.instance_transforms.inference}
|
| 201 |
+
sim_psf_config: ${psf}
|
| 202 |
+
dataloader:
|
| 203 |
+
train:
|
| 204 |
+
_target_: torch.utils.data.DataLoader
|
| 205 |
+
batch_size: 1
|
| 206 |
+
num_workers: 2
|
| 207 |
+
pin_memory: true
|
| 208 |
+
inference:
|
| 209 |
+
_target_: torch.utils.data.DataLoader
|
| 210 |
+
batch_size: 1
|
| 211 |
+
num_workers: 2
|
| 212 |
+
pin_memory: true
|
| 213 |
+
transforms:
|
| 214 |
+
instance_transforms:
|
| 215 |
+
train:
|
| 216 |
+
all:
|
| 217 |
+
_target_: torchvision.transforms.v2.Compose
|
| 218 |
+
transforms:
|
| 219 |
+
- _target_: src.transforms.PadCrop
|
| 220 |
+
length: 4
|
| 221 |
+
pad_format: replicated
|
| 222 |
+
random_crop: true
|
| 223 |
+
inference: null
|
| 224 |
+
batch_transforms:
|
| 225 |
+
train: null
|
| 226 |
+
inference: null
|
| 227 |
+
codec:
|
| 228 |
+
_target_: src.transforms.CodecEncoderDecoder
|
| 229 |
+
codec_cls: ${resolve_class:dac.DAC}
|
| 230 |
+
codec_weights_path: data/dac_exps/${codec.codec_name}/latest/dac/weights.pth
|
| 231 |
+
codec_add_root_path: true
|
| 232 |
+
codec_kwargs: null
|
| 233 |
+
codec_name: 32x32_120_16khz_original
|
| 234 |
+
eval_mode: true
|
| 235 |
+
freeze_weights: true
|
| 236 |
+
reconstruction:
|
| 237 |
+
roi_kwargs:
|
| 238 |
+
top_left:
|
| 239 |
+
- 65
|
| 240 |
+
- 118
|
| 241 |
+
height: 256
|
| 242 |
+
width: 256
|
| 243 |
+
resize_coef: 8
|
| 244 |
+
normalize_lensless: true
|
| 245 |
+
corners_list: null
|
| 246 |
+
psf:
|
| 247 |
+
slm: adafruit
|
| 248 |
+
sensor: rpi_hq
|
| 249 |
+
downsample: 8
|
| 250 |
+
rotate: -0.8
|
| 251 |
+
vertical_shift: -20
|
| 252 |
+
horizontal_shift: -20
|
| 253 |
+
flipud: true
|
| 254 |
+
use_waveprop: true
|
| 255 |
+
deadspace: true
|
| 256 |
+
scene2mask: 0.3
|
| 257 |
+
mask2sensor: 0.004
|
| 258 |
+
grayscale: true
|
| 259 |
+
lr_scheduler:
|
| 260 |
+
_target_: torch.optim.lr_scheduler.ConstantLR
|
| 261 |
+
factor: 1
|
| 262 |
+
optimizer:
|
| 263 |
+
_target_: torch.optim.Adam
|
| 264 |
+
lr: 0.0001
|
| 265 |
+
loss_function:
|
| 266 |
+
_target_: src.loss.ReconstructionLoss
|
| 267 |
+
codec_mse_coef: 1
|
| 268 |
+
codec_ssim_coef: 1
|
| 269 |
+
codec_gmsd_coef: 0
|
| 270 |
+
raw_codec_ssim_coef: 1
|
| 271 |
+
raw_codec_l1_coef: 0
|
| 272 |
+
audio_l1_coef: 0
|
| 273 |
+
audio_sisdr_coef: 0
|
| 274 |
+
audio_stft_coef: 0
|
| 275 |
+
audio_mel_coef: 0.1
|
| 276 |
+
audio_stft_config:
|
| 277 |
+
window_lengths:
|
| 278 |
+
- 2048
|
| 279 |
+
- 512
|
| 280 |
+
audio_mel_config:
|
| 281 |
+
n_mels:
|
| 282 |
+
- 5
|
| 283 |
+
- 10
|
| 284 |
+
- 20
|
| 285 |
+
- 40
|
| 286 |
+
window_lengths:
|
| 287 |
+
- 32
|
| 288 |
+
- 64
|
| 289 |
+
- 128
|
| 290 |
+
- 256
|
| 291 |
+
mel_fmin:
|
| 292 |
+
- 0
|
| 293 |
+
- 0
|
| 294 |
+
- 0
|
| 295 |
+
- 0
|
| 296 |
+
mel_fmax:
|
| 297 |
+
- null
|
| 298 |
+
- null
|
| 299 |
+
- null
|
| 300 |
+
- null
|
| 301 |
+
pow: 1.0
|
| 302 |
+
clamp_eps: 1.0e-05
|
| 303 |
+
mag_weight: 0.0
|
| 304 |
+
resize_coef: ${reconstruction.resize_coef}
|
| 305 |
+
ssim_kernel: 7
|
| 306 |
+
ssim_sigma: 0.5
|
| 307 |
+
raw_ssim_kernel: 11
|
| 308 |
+
trainer:
|
| 309 |
+
log_step: 50
|
| 310 |
+
n_epochs: 100
|
| 311 |
+
epoch_len: 500
|
| 312 |
+
device_tensors:
|
| 313 |
+
- lensless_codec_video
|
| 314 |
+
- lensed_codec_video
|
| 315 |
+
- lensless_psf
|
| 316 |
+
- audio
|
| 317 |
+
resume_from: null
|
| 318 |
+
device: auto
|
| 319 |
+
override: true
|
| 320 |
+
monitor: max test_PSNR
|
| 321 |
+
save_period: 5
|
| 322 |
+
early_stop: ${trainer.n_epochs}
|
| 323 |
+
save_dir: saved
|
| 324 |
+
seed: 1
|
32x32_random_mse_ssim_raw_ssim_PSF_Unet4M_U5_Unet4M/checkpoint-epoch100.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a894df0b3866eeecd507bedceefaf8a381d7644be1b5b1877321ac4b31c39ce1
|
| 3 |
+
size 97416570
|
32x32_random_mse_ssim_raw_ssim_PSF_Unet4M_U5_Unet4M/config.yaml
ADDED
|
@@ -0,0 +1,314 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
model:
|
| 2 |
+
_target_: src.model.LenslessWrapper
|
| 3 |
+
use_loader: false
|
| 4 |
+
loader_kwargs: null
|
| 5 |
+
use_batch_video_version: false
|
| 6 |
+
freeze_weights: false
|
| 7 |
+
psf_path: data/digicam_psf/SIM_psf.png
|
| 8 |
+
psf_loader_kwargs:
|
| 9 |
+
downsample: 8
|
| 10 |
+
return_bg: false
|
| 11 |
+
grayscale_psf: true
|
| 12 |
+
recon_name: UnrolledADMM
|
| 13 |
+
recon_kwargs:
|
| 14 |
+
n_iter: 5
|
| 15 |
+
mu1: 0.0001
|
| 16 |
+
mu2: 0.0001
|
| 17 |
+
mu3: 0.0001
|
| 18 |
+
tau: 0.0002
|
| 19 |
+
pre_process:
|
| 20 |
+
_target_: lensless.recon.drunet.network_unet.UNetRes
|
| 21 |
+
in_nc: 2
|
| 22 |
+
out_nc: 1
|
| 23 |
+
nc:
|
| 24 |
+
- 32
|
| 25 |
+
- 64
|
| 26 |
+
- 112
|
| 27 |
+
- 128
|
| 28 |
+
nb: 4
|
| 29 |
+
act_mode: R
|
| 30 |
+
downsample_mode: strideconv
|
| 31 |
+
upsample_mode: convtranspose
|
| 32 |
+
post_process:
|
| 33 |
+
_target_: lensless.recon.drunet.network_unet.UNetRes
|
| 34 |
+
in_nc: 2
|
| 35 |
+
out_nc: 1
|
| 36 |
+
nc:
|
| 37 |
+
- 32
|
| 38 |
+
- 64
|
| 39 |
+
- 116
|
| 40 |
+
- 128
|
| 41 |
+
nb: 4
|
| 42 |
+
act_mode: R
|
| 43 |
+
downsample_mode: strideconv
|
| 44 |
+
upsample_mode: convtranspose
|
| 45 |
+
psf_network:
|
| 46 |
+
_target_: lensless.recon.drunet.network_unet.UNetRes
|
| 47 |
+
in_nc: 2
|
| 48 |
+
out_nc: 1
|
| 49 |
+
nc:
|
| 50 |
+
- 4
|
| 51 |
+
- 8
|
| 52 |
+
- 16
|
| 53 |
+
- 32
|
| 54 |
+
nb: 4
|
| 55 |
+
act_mode: R
|
| 56 |
+
downsample_mode: strideconv
|
| 57 |
+
upsample_mode: convtranspose
|
| 58 |
+
psf_residual: false
|
| 59 |
+
skip_unrolled: false
|
| 60 |
+
return_intermediate: false
|
| 61 |
+
writer:
|
| 62 |
+
_target_: src.logger.WandBWriter
|
| 63 |
+
project_name: lenslessmic
|
| 64 |
+
entity: null
|
| 65 |
+
run_name: 32x32_random_mse_ssim_raw_ssim_PSF_Unet4M_U5_Unet4M
|
| 66 |
+
mode: online
|
| 67 |
+
loss_names:
|
| 68 |
+
- loss
|
| 69 |
+
- codec_mse_loss
|
| 70 |
+
- codec_ssim_loss
|
| 71 |
+
- raw_codec_ssim_loss
|
| 72 |
+
- raw_codec_l1_loss
|
| 73 |
+
- audio_l1_loss
|
| 74 |
+
- audio_sisdr_loss
|
| 75 |
+
- audio_stft_loss
|
| 76 |
+
- audio_mel_loss
|
| 77 |
+
log_checkpoints: false
|
| 78 |
+
id_length: 8
|
| 79 |
+
names:
|
| 80 |
+
- input-1:frame
|
| 81 |
+
- input-2:frame
|
| 82 |
+
- input-3:frame
|
| 83 |
+
- input-4:frame
|
| 84 |
+
figsize:
|
| 85 |
+
- 15
|
| 86 |
+
- 15
|
| 87 |
+
sample_rate: 16000
|
| 88 |
+
run_id: u0zp8tee
|
| 89 |
+
metrics:
|
| 90 |
+
device: auto
|
| 91 |
+
train:
|
| 92 |
+
- _target_: src.metrics.SISDRMetric
|
| 93 |
+
name: SISDR
|
| 94 |
+
- _target_: src.metrics.PSNRMetric
|
| 95 |
+
name: PSNR
|
| 96 |
+
- _target_: src.metrics.QuantizationMatchMetric
|
| 97 |
+
name: QuantizationMatch-all
|
| 98 |
+
codebook_index: all
|
| 99 |
+
inference:
|
| 100 |
+
- _target_: src.metrics.SISDRMetric
|
| 101 |
+
name: SISDR
|
| 102 |
+
- _target_: src.metrics.STOIMetric
|
| 103 |
+
name: STOI
|
| 104 |
+
- _target_: src.metrics.WERMetric
|
| 105 |
+
name: WER
|
| 106 |
+
- _target_: src.metrics.PESQMetric
|
| 107 |
+
name: PESQ
|
| 108 |
+
- _target_: src.metrics.MelMetric
|
| 109 |
+
name: Mel
|
| 110 |
+
audio_mel_config:
|
| 111 |
+
n_mels:
|
| 112 |
+
- 5
|
| 113 |
+
- 10
|
| 114 |
+
- 20
|
| 115 |
+
- 40
|
| 116 |
+
- 80
|
| 117 |
+
- 160
|
| 118 |
+
- 320
|
| 119 |
+
window_lengths:
|
| 120 |
+
- 32
|
| 121 |
+
- 64
|
| 122 |
+
- 128
|
| 123 |
+
- 256
|
| 124 |
+
- 512
|
| 125 |
+
- 1024
|
| 126 |
+
- 2048
|
| 127 |
+
mel_fmin:
|
| 128 |
+
- 0
|
| 129 |
+
- 0
|
| 130 |
+
- 0
|
| 131 |
+
- 0
|
| 132 |
+
- 0
|
| 133 |
+
- 0
|
| 134 |
+
- 0
|
| 135 |
+
mel_fmax:
|
| 136 |
+
- null
|
| 137 |
+
- null
|
| 138 |
+
- null
|
| 139 |
+
- null
|
| 140 |
+
- null
|
| 141 |
+
- null
|
| 142 |
+
- null
|
| 143 |
+
pow: 1.0
|
| 144 |
+
clamp_eps: 1.0e-05
|
| 145 |
+
mag_weight: 0.0
|
| 146 |
+
- _target_: src.metrics.STFTMetric
|
| 147 |
+
name: STFT
|
| 148 |
+
audio_stft_config:
|
| 149 |
+
window_lengths:
|
| 150 |
+
- 2048
|
| 151 |
+
- 512
|
| 152 |
+
- _target_: src.metrics.QuantizationMatchMetric
|
| 153 |
+
name: QuantizationMatch-all
|
| 154 |
+
codebook_index: all
|
| 155 |
+
- _target_: src.metrics.QuantizationMatchMetric
|
| 156 |
+
name: QuantizationMatch-1
|
| 157 |
+
codebook_index: 1
|
| 158 |
+
- _target_: src.metrics.QuantizationMatchMetric
|
| 159 |
+
name: QuantizationMatch-2
|
| 160 |
+
codebook_index: 2
|
| 161 |
+
- _target_: src.metrics.PSNRMetric
|
| 162 |
+
name: PSNR
|
| 163 |
+
- _target_: src.metrics.SSIMMetric
|
| 164 |
+
name: SSIM
|
| 165 |
+
- _target_: src.metrics.GMSDMetric
|
| 166 |
+
name: GMSD
|
| 167 |
+
- _target_: src.metrics.MSEMetric
|
| 168 |
+
name: MSE
|
| 169 |
+
normalized: false
|
| 170 |
+
- _target_: src.metrics.MSEMetric
|
| 171 |
+
name: NormMSE
|
| 172 |
+
normalized: true
|
| 173 |
+
datasets:
|
| 174 |
+
train:
|
| 175 |
+
_target_: src.datasets.RandomDataset
|
| 176 |
+
part: train
|
| 177 |
+
roi_kwargs: ${reconstruction.roi_kwargs}
|
| 178 |
+
codec_name: ${codec.codec_name}
|
| 179 |
+
lensless_tag: measurement
|
| 180 |
+
instance_transforms: ${transforms.instance_transforms.train}
|
| 181 |
+
sim_psf_config: ${psf}
|
| 182 |
+
test:
|
| 183 |
+
_target_: src.datasets.LibrispeechDataset
|
| 184 |
+
limit: 1
|
| 185 |
+
max_audio_length: 3
|
| 186 |
+
part: test-clean
|
| 187 |
+
roi_kwargs: ${reconstruction.roi_kwargs}
|
| 188 |
+
codec_name: ${codec.codec_name}
|
| 189 |
+
lensless_tag: measurement
|
| 190 |
+
instance_transforms: ${transforms.instance_transforms.inference}
|
| 191 |
+
sim_psf_config: ${psf}
|
| 192 |
+
dataloader:
|
| 193 |
+
train:
|
| 194 |
+
_target_: torch.utils.data.DataLoader
|
| 195 |
+
batch_size: 1
|
| 196 |
+
num_workers: 2
|
| 197 |
+
pin_memory: true
|
| 198 |
+
inference:
|
| 199 |
+
_target_: torch.utils.data.DataLoader
|
| 200 |
+
batch_size: 1
|
| 201 |
+
num_workers: 2
|
| 202 |
+
pin_memory: true
|
| 203 |
+
transforms:
|
| 204 |
+
instance_transforms:
|
| 205 |
+
train:
|
| 206 |
+
all:
|
| 207 |
+
_target_: torchvision.transforms.v2.Compose
|
| 208 |
+
transforms:
|
| 209 |
+
- _target_: src.transforms.PadCrop
|
| 210 |
+
length: 4
|
| 211 |
+
pad_format: replicated
|
| 212 |
+
random_crop: true
|
| 213 |
+
inference: null
|
| 214 |
+
batch_transforms:
|
| 215 |
+
train: null
|
| 216 |
+
inference: null
|
| 217 |
+
codec:
|
| 218 |
+
_target_: src.transforms.CodecEncoderDecoder
|
| 219 |
+
codec_cls: ${resolve_class:dac.DAC}
|
| 220 |
+
codec_weights_path: data/dac_exps/${codec.codec_name}/latest/dac/weights.pth
|
| 221 |
+
codec_add_root_path: true
|
| 222 |
+
codec_kwargs: null
|
| 223 |
+
codec_name: 32x32_120_16khz_original
|
| 224 |
+
eval_mode: true
|
| 225 |
+
freeze_weights: true
|
| 226 |
+
reconstruction:
|
| 227 |
+
roi_kwargs:
|
| 228 |
+
top_left:
|
| 229 |
+
- 65
|
| 230 |
+
- 118
|
| 231 |
+
height: 256
|
| 232 |
+
width: 256
|
| 233 |
+
resize_coef: 8
|
| 234 |
+
normalize_lensless: true
|
| 235 |
+
corners_list: null
|
| 236 |
+
psf:
|
| 237 |
+
slm: adafruit
|
| 238 |
+
sensor: rpi_hq
|
| 239 |
+
downsample: 8
|
| 240 |
+
rotate: -0.8
|
| 241 |
+
vertical_shift: -20
|
| 242 |
+
horizontal_shift: -20
|
| 243 |
+
flipud: true
|
| 244 |
+
use_waveprop: true
|
| 245 |
+
deadspace: true
|
| 246 |
+
scene2mask: 0.3
|
| 247 |
+
mask2sensor: 0.004
|
| 248 |
+
grayscale: true
|
| 249 |
+
lr_scheduler:
|
| 250 |
+
_target_: torch.optim.lr_scheduler.ConstantLR
|
| 251 |
+
factor: 1
|
| 252 |
+
optimizer:
|
| 253 |
+
_target_: torch.optim.Adam
|
| 254 |
+
lr: 0.0001
|
| 255 |
+
loss_function:
|
| 256 |
+
_target_: src.loss.ReconstructionLoss
|
| 257 |
+
codec_mse_coef: 1
|
| 258 |
+
codec_ssim_coef: 1
|
| 259 |
+
codec_gmsd_coef: 0
|
| 260 |
+
raw_codec_ssim_coef: 1
|
| 261 |
+
raw_codec_l1_coef: 0
|
| 262 |
+
audio_l1_coef: 0
|
| 263 |
+
audio_sisdr_coef: 0
|
| 264 |
+
audio_stft_coef: 0
|
| 265 |
+
audio_mel_coef: 0
|
| 266 |
+
audio_stft_config:
|
| 267 |
+
window_lengths:
|
| 268 |
+
- 2048
|
| 269 |
+
- 512
|
| 270 |
+
audio_mel_config:
|
| 271 |
+
n_mels:
|
| 272 |
+
- 5
|
| 273 |
+
- 10
|
| 274 |
+
- 20
|
| 275 |
+
- 40
|
| 276 |
+
window_lengths:
|
| 277 |
+
- 32
|
| 278 |
+
- 64
|
| 279 |
+
- 128
|
| 280 |
+
- 256
|
| 281 |
+
mel_fmin:
|
| 282 |
+
- 0
|
| 283 |
+
- 0
|
| 284 |
+
- 0
|
| 285 |
+
- 0
|
| 286 |
+
mel_fmax:
|
| 287 |
+
- null
|
| 288 |
+
- null
|
| 289 |
+
- null
|
| 290 |
+
- null
|
| 291 |
+
pow: 1.0
|
| 292 |
+
clamp_eps: 1.0e-05
|
| 293 |
+
mag_weight: 0.0
|
| 294 |
+
resize_coef: ${reconstruction.resize_coef}
|
| 295 |
+
ssim_kernel: 7
|
| 296 |
+
ssim_sigma: 0.5
|
| 297 |
+
raw_ssim_kernel: 11
|
| 298 |
+
trainer:
|
| 299 |
+
log_step: 50
|
| 300 |
+
n_epochs: 100
|
| 301 |
+
epoch_len: 500
|
| 302 |
+
device_tensors:
|
| 303 |
+
- lensless_codec_video
|
| 304 |
+
- lensed_codec_video
|
| 305 |
+
- lensless_psf
|
| 306 |
+
- audio
|
| 307 |
+
resume_from: null
|
| 308 |
+
device: auto
|
| 309 |
+
override: true
|
| 310 |
+
monitor: max test_PSNR
|
| 311 |
+
save_period: 5
|
| 312 |
+
early_stop: ${trainer.n_epochs}
|
| 313 |
+
save_dir: saved
|
| 314 |
+
seed: 1
|