"simpler/train0.hdf5",
"simpler/train1.hdf5",
"simpler/train2.hdf5",
"simpler/train3.hdf5",
"simpler/train4.hdf5",
"simpler/train5.hdf5",
)
dynamics_trainset = omnipush_datasets.OmnipushDynamicsDataset(
*omnipush_train_files, **dataset_args)
dynamics_recurrent_trainset = omnipush_datasets.OmnipushSubsequenceDataset(
*omnipush_train_files, subsequence_length=16, **dataset_args)
measurement_trainset = omnipush_datasets.OmnipushMeasurementDataset(
*omnipush_train_files, samples_per_pair=10, **dataset_args)
e2e_trainset = omnipush_datasets.OmnipushParticleFilterDataset(
*omnipush_train_files,
subsequence_length=16,
particle_count=30,
particle_stddev=(.1, .1),
**dataset_args)
# Pre-train measurement
models = [
(pf_image_model, 'measurement_image'),
# (pf_force_model, 'measurement_force'),
]
measurement_trainset_loader = torch.utils.data.DataLoader(measurement_trainset,
batch_size=32,
shuffle=True,
num_workers=16)
for pf_model, optim_name in models:
for i in range(MEASUREMENT_PRETRAIN_EPOCHS):
print("Pre-training measurement epoch", i)
buddy.load_checkpoint(path=args.load_checkpoint)
if args.module_type == "ekf":
buddy.load_checkpoint_module(source="image_model",
path=args.load_checkpoint)
buddy.load_checkpoint_module(source="force_model",
path=args.load_checkpoint)
print("Creating dataset...")
if args.omnipush:
e2e_trainset = omnipush_datasets.OmnipushParticleFilterDataset(
"simpler/train0.hdf5",
"simpler/train1.hdf5",
"simpler/train2.hdf5",
"simpler/train3.hdf5",
"simpler/train4.hdf5",
"simpler/train5.hdf5",
subsequence_length=16,
particle_count=1,
particle_stddev=(.03, .03),
**dataset_args)
dataset_measurement = omnipush_datasets.OmnipushMeasurementDataset(
"simpler/train0.hdf5",
"simpler/train1.hdf5",
"simpler/train2.hdf5",
"simpler/train3.hdf5",
"simpler/train4.hdf5",
"simpler/train5.hdf5",
subsequence_length=16,
stddev=(0.5, 0.5),