def main():
""" """
pyplot.style.use("bmh")
base_path = Path.home() / "Data" / "Datasets" / "Clouds"
resized_loc = base_path / "resized"
save_model_path = PROJECT_APP_PATH.user_data / "cloud_seg.model"
SEED = 87539842
batch_size = 8
num_workers = 0
seed_stack(SEED)
min_size = (10000, 10000, 10000, 10000)
train_loader = DataLoader(
CloudSegmentationDataset(
df_path=base_path / "train.csv",
resized_loc=resized_loc,
subset=SplitEnum.training,
),
batch_size=batch_size,
shuffle=True,
num_workers=num_workers,
)
valid_loader = DataLoader(
CloudSegmentationDataset(
df_path=base_path / "train.csv",
resized_loc=resized_loc,
subset=SplitEnum.validation,
),
batch_size=batch_size,
shuffle=False,
num_workers=num_workers,
)
model = SkipHourglassFission(
CloudSegmentationDataset.predictors_shape[-1],
CloudSegmentationDataset.response_shape,
encoding_depth=2,
)
model.to(global_torch_device())
if save_model_path.exists():
model.load_state_dict(torch.load(
str(save_model_path))) # load last model
criterion = BCEDiceLoss(eps=1.0)
optimiser = torch.optim.SGD(model.parameters(), lr=0.01)
current_lr = next(iter(optimiser.param_groups))["lr"]
scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(optimiser,
7,
eta_min=current_lr /
100,
last_epoch=-1)
model = train_d(
model,
train_loader,
valid_loader,
criterion,
optimiser,
scheduler,
str(save_model_path),
)
if save_model_path.exists():
model.load_state_dict(torch.load(
str(save_model_path))) # load best model
model.eval()
# %%
valid_masks = []
count = 0
tr = min(len(valid_loader.dataset) * 4, 2000)
probabilities = numpy.zeros((tr, 350, 525), dtype=numpy.float32)
for data, target in tqdm(valid_loader):
data = data.to(global_torch_device())
target = target.cpu().detach().numpy()
outpu, *_ = model(data)
outpu = torch.sigmoid(outpu).cpu().detach().numpy()
for p in range(data.shape[0]):
output, mask = outpu[p], target[p]
for m in mask:
valid_masks.append(resize_image_cv(m))
for probability in output:
probabilities[count, :, :] = resize_image_cv(probability)
count += 1
if count >= tr - 1:
break
if count >= tr - 1:
break
class_parameters = grid_search(model, probabilities, valid_masks,
valid_loader)
submission(
model,
class_parameters,
base_path=base_path,
batch_size=batch_size,
resized_loc=resized_loc,
)
def main(
base_path: Path = Path.home() / "Data" / "Datasets" / "PennFudanPed",
train_model: bool = True,
load_prev_model: bool = True,
writer: Writer = TensorBoardPytorchWriter(PROJECT_APP_PATH.user_log /
"instanced_person_segmentation" /
f"{time.time()}"),
):
""" """
# base_path = Path("/") / "encrypted_disk" / "heider" / "Data" / "PennFudanPed"
base_path: Path = Path.home() / "Data3" / "PennFudanPed"
# base_path = Path('/media/heider/OS/Users/Christian/Data/Datasets/') / "PennFudanPed"
pyplot.style.use("bmh")
save_model_path = (
ensure_existence(PROJECT_APP_PATH.user_data / "models") /
"instanced_penn_fudan_ped_seg.model")
eval_model = not train_model
SEED = 9221
batch_size = 32
num_workers = 0
encoding_depth = 2
learning_rate = 6e-6 # sequence 6e-2 6e-3 6e-4 6e-5
seed_stack(SEED)
train_set = PennFudanDataset(
base_path,
SplitEnum.training,
return_variant=PennFudanDataset.PennFudanReturnVariantEnum.instanced,
)
train_loader = DataLoader(train_set,
batch_size=batch_size,
shuffle=True,
num_workers=num_workers)
valid_loader = DataLoader(
PennFudanDataset(
base_path,
SplitEnum.validation,
return_variant=PennFudanDataset.PennFudanReturnVariantEnum.
instanced,
),
batch_size=batch_size,
shuffle=False,
num_workers=num_workers,
)
model = SkipHourglassFission(
input_channels=train_set.predictor_shape[-1],
output_heads=(train_set.response_shape[-1], ),
encoding_depth=encoding_depth,
)
model.to(global_torch_device())
if load_prev_model and save_model_path.exists():
model.load_state_dict(torch.load(str(save_model_path)))
print("loading saved model")
if train_model:
with TorchTrainSession(model):
criterion = BCEDiceLoss()
# optimiser = torch.optim.SGD(model.parameters(), lr=learning_rate)
optimiser = torch.optim.Adam(model.parameters(), lr=learning_rate)
# scheduler = torch.optim.lr_scheduler.CosineAnnealingLR( optimiser, T_max=7, eta_min=learning_rate / 100, last_epoch=-1 )
model = train_person_segmentor(
model,
train_loader,
valid_loader,
criterion,
optimiser,
save_model_path=save_model_path,
learning_rate=learning_rate,
writer=writer,
)
if eval_model:
validate_model(model, valid_loader)
def main(model_name: str = "maskrcnn_pennfudanped", score_threshold=0.55):
"""
Args:
model_name:
score_threshold:
"""
base_path = PROJECT_APP_PATH.user_data / "maskrcnn"
dataset_root = Path.home() / "Data"
seed_stack(3825)
dataset = (
PennFudanDataset # (dataset_root / "PennFudanPed", SplitEnum.training)
)
categories = dataset.categories
if True:
model = load_model(model_name=model_name,
model_directory=base_path / "models")
else:
model = get_pretrained_instance_segmentation_maskrcnn(
dataset.response_channels)
model.to(global_torch_device())
cpu_device = torch.device("cpu")
with torch.no_grad():
with TorchEvalSession(model):
for image in tqdm(
to_tensor_generator(
frame_generator(cv2.VideoCapture(0)),
device=global_torch_device(),
)):
prediction = model(
# torch_vision_normalize_batch_nchw(
uint_hwc_to_chw_float_tensor(image).unsqueeze(0)
# )
)[0]
(boxes, labels, scores) = (
prediction["boxes"].to(cpu_device).numpy(),
prediction["labels"].to(cpu_device).numpy(),
torch.sigmoid(
prediction["scores"]).to(cpu_device).numpy(),
)
indices = scores > score_threshold
if show_image(
draw_bounding_boxes(
quick_to_pil_image(image),
boxes[indices],
labels=labels[indices],
scores=scores[indices],
categories=categories,
),
model_name,
wait=True,
):
break # esc to quit
def export_detection_model(
model_export_path: Path = ensure_existence(
PROJECT_APP_PATH.user_data / "penn_fudan_segmentation"
)
/ "seg_skip_fis",
SEED: int = 87539842,
) -> None:
"""
:param model_export_path:
:type model_export_path:
:return:
:rtype:"""
model = OutputActivationModule(
SkipHourglassFission(input_channels=3, output_heads=(1,), encoding_depth=1)
)
with TorchDeviceSession(device=global_torch_device("cpu"), model=model):
with TorchEvalSession(model):
seed_stack(SEED)
# standard PyTorch mean-std input image normalization
transform = transforms.Compose(
[
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
]
)
frame_g = frame_generator(cv2.VideoCapture(0))
for image in tqdm(frame_g):
example_input = (
transform(image).unsqueeze(0).to(global_torch_device()),
)
try:
traced_script_module = torch.jit.trace(
model,
example_input,
# strict=strict_jit,
check_inputs=(
transform(next(frame_g))
.unsqueeze(0)
.to(global_torch_device()),
transform(next(frame_g))
.unsqueeze(0)
.to(global_torch_device()),
),
)
exp_path = model_export_path.with_suffix(".traced")
traced_script_module.save(str(exp_path))
print(
f"Traced Ops used {torch.jit.export_opnames(traced_script_module)}"
)
sprint(
f"Successfully exported JIT Traced model at {exp_path}",
color="green",
)
except Exception as e_i:
sprint(f"Torch JIT Trace export does not work!, {e_i}", color="red")
break
def main(
dataset_root=Path.home() / "DataWin" / "Datasets",
model_name=f"maskrcnn_pennfudanped",
):
""" """
base_path = ensure_existence(PROJECT_APP_PATH.user_data / "maskrcnn")
log_path = ensure_existence(PROJECT_APP_PATH.user_log / "maskrcnn")
export_root = ensure_existence(base_path / "models")
batch_size = 4
num_epochs = 10
optimiser_spec = GDKC(torch.optim.Adam, lr=3e-4)
scheduler_spec = GDKC(
torch.optim.lr_scheduler.
StepLR, # a learning rate scheduler which decreases the learning rate by
step_size=3, # 10x every 3 epochs
gamma=0.1,
)
num_workers = 0
seed_stack(3825)
dataset = PennFudanDataset(
dataset_root / "PennFudanPed",
SplitEnum.training,
return_variant=PennFudanDataset.PennFudanReturnVariantEnum.all,
)
dataset_validation = PennFudanDataset(
dataset_root / "PennFudanPed",
SplitEnum.validation,
return_variant=PennFudanDataset.PennFudanReturnVariantEnum.all,
)
split = SplitIndexer(len(dataset), validation=0.3, testing=0)
split_indices = torch.randperm(split.total_num).tolist()
data_loader = DataLoader(
Subset(dataset, split_indices[:-split.validation_num]),
batch_size=batch_size,
shuffle=True,
num_workers=num_workers,
collate_fn=collate_first_dim,
)
data_loader_val = DataLoader(
Subset(dataset_validation, split_indices[-split.validation_num:]),
batch_size=1,
shuffle=False,
num_workers=num_workers,
collate_fn=collate_first_dim,
)
model = get_pretrained_instance_segmentation_maskrcnn(
dataset.response_channels)
optimiser = optimiser_spec(trainable_parameters(model))
lr_scheduler = scheduler_spec(optimiser)
if True:
candidate = load_model(
model_name=model_name,
model_directory=export_root,
raise_on_failure=False,
)
if candidate:
model = candidate
if True:
with TorchTrainSession(model):
with TensorBoardPytorchWriter(log_path / model_name) as writer:
for epoch_i in tqdm(range(num_epochs), desc="Epoch #"):
maskrcnn_train_single_epoch(
model=model,
optimiser=optimiser,
data_loader=data_loader,
writer=writer,
)
lr_scheduler.step() # update the learning rate
maskrcnn_evaluate(
model, data_loader_val, writer=writer
) # evaluate on the validation dataset
save_model(model,
model_name=model_name,
save_directory=export_root)
if True:
with TorchEvalSession(model): # put the model in evaluation mode
img, _ = dataset_validation[
0] # pick one image from the test set
with torch.no_grad():
prediction = model([img.to(global_torch_device())])
from matplotlib import pyplot
pyplot.imshow(
Image.fromarray(
img.mul(255).permute(1, 2, 0).byte().numpy()))
pyplot.show()
import cv2
pyplot.imshow(
Image.fromarray(prediction[0]["masks"][0, 0].mul(
255).byte().cpu().numpy()))
pyplot.show()
(boxes, labels, scores) = (
prediction[0]["boxes"].to("cpu").numpy(),
prediction[0]["labels"].to("cpu").numpy(),
torch.sigmoid(prediction[0]["scores"]).to("cpu").numpy(),
)
from draugr.opencv_utilities import draw_bounding_boxes
from draugr.torch_utilities.images.conversion import quick_to_pil_image
indices = scores > 0.1
cv2.namedWindow(model_name, cv2.WINDOW_NORMAL)
cv2.imshow(
model_name,
draw_bounding_boxes(
quick_to_pil_image(img),
boxes[indices],
labels=labels[indices],
scores=scores[indices],
# categories=categories,
),
)
cv2.waitKey()
def training_procedure(n_epoch=60,
train: bool = True,
load_previous: bool = True):
train_set = SpeechCommands(split=Split.Training)
waveform, sample_rate, category, speaker_id, utterance_number = train_set[
0]
transform = torch.nn.Sequential(
torchaudio.transforms.Resample(orig_freq=sample_rate,
new_freq=new_sample_rate))
model = M5(n_input=transform(waveform).shape[0],
n_output=len(train_set.categories))
persistence_model_name = f'{model_name}_{get_model_hash(transform)}_{get_model_hash(model)}'
if load_previous:
candidate = load_model(model_name=persistence_model_name,
model_directory=model_storage_path,
raise_on_failure=False)
if candidate:
model = candidate
transform.to(device)
model.to(device)
collate_fn = collate_transform_wrapped(train_set.label_to_index, transform)
# print(f"Trainable parameters: {named_trainable_parameters(model)}")
print(
f"Number of trainable parameters: {get_num_parameters(model, only_trainable=True)}"
)
if train:
train_loader = DataLoader(
train_set,
batch_size=batch_size,
shuffle=True,
drop_last=drop_last_train,
collate_fn=collate_fn,
num_workers=num_workers,
pin_memory=pin_memory,
)
valid_loader = DataLoader(
SpeechCommands(split=Split.Validation),
batch_size=batch_size,
shuffle=False,
drop_last=False,
collate_fn=collate_fn,
num_workers=num_workers,
pin_memory=pin_memory,
)
optimiser = optim.Adam(trainable_parameters(model),
lr=0.01,
weight_decay=0.0001)
scheduler = optim.lr_scheduler.StepLR(
optimiser, step_size=20,
gamma=0.1) # reduce the learning after 20 epochs by a factor of 10
with TensorBoardPytorchWriter(PROJECT_APP_PATH.user_log / model_name /
str(time.time())) as writer:
best_valid_acc = math.inf
for epoch in tqdm(range(1, n_epoch + 1),
total=n_epoch,
desc='Epoch #'):
single_epoch_fitting(model,
optimiser,
train_loader,
epoch=epoch,
writer=writer,
device_=device)
scheduler.step()
acc = single_epoch_evaluation(model,
valid_loader,
subset=Split.Validation,
epoch=epoch,
writer=writer,
device=device)
if acc < best_valid_acc:
best_valid_acc = acc
save_model(model,
model_name=persistence_model_name,
save_directory=model_storage_path)
writer.blip('new_best_model', epoch)
else:
seed_stack(0)
with TensorBoardPytorchWriter(PROJECT_APP_PATH.user_log / model_name /
str(time.time())) as writer:
test_loader = DataLoader(
SpeechCommands(split=Split.Testing),
batch_size=1,
shuffle=True,
drop_last=False,
collate_fn=collate_fn,
num_workers=num_workers,
pin_memory=pin_memory,
)
acc = single_epoch_evaluation(model,
test_loader,
subset=Split.Testing,
epoch=0,
writer=writer,
device=device)
print(acc)
find_n_misclassified(model,
test_loader,
mapper=train_set.index_to_label)
from neodroidaudition.classification import M5
from neodroidaudition.data import SpeechCommands
from neodroidaudition.torch_utilities.collation import collate_transform_wrapped
model_name, model_storage_path = 'm5_speech_command', PROJECT_APP_PATH.user_data / 'speech_command' / 'models'
device = global_torch_device(True)
if device == "cuda":
num_workers = 1
pin_memory = True
else:
num_workers = 0
pin_memory = False
drop_last_train = False
seed_stack(0)
batch_size = 256
new_sample_rate = 8000
def training_procedure(n_epoch=60,
train: bool = True,
load_previous: bool = True):
train_set = SpeechCommands(split=Split.Training)
waveform, sample_rate, category, speaker_id, utterance_number = train_set[
0]
transform = torch.nn.Sequential(
torchaudio.transforms.Resample(orig_freq=sample_rate,
new_freq=new_sample_rate))
def main():
""" """
pyplot.style.use("bmh")
base_dataset_path = Path.home() / "Data" / "Datasets" / "Clouds"
image_path = base_dataset_path / "resized"
save_model_path = PROJECT_APP_PATH.user_data / "cloud_seg.model"
SEED = 87539842
batch_size = 8
num_workers = 0
seed_stack(SEED)
train_loader = DataLoader(
CloudSegmentationDataset(base_dataset_path,
image_path,
subset=SplitEnum.training),
batch_size=batch_size,
shuffle=True,
num_workers=num_workers,
)
valid_loader = DataLoader(
CloudSegmentationDataset(base_dataset_path,
image_path,
subset=SplitEnum.validation),
batch_size=batch_size,
shuffle=False,
num_workers=num_workers,
)
test_loader = DataLoader(
CloudSegmentationDataset(base_dataset_path,
image_path,
subset=SplitEnum.testing),
batch_size=batch_size,
shuffle=False,
num_workers=num_workers,
)
model = SkipHourglassFission(
CloudSegmentationDataset.predictor_channels,
(CloudSegmentationDataset.response_channels, ),
encoding_depth=1,
)
model.to(global_torch_device())
if save_model_path.exists():
model.load_state_dict(torch.load(
str(save_model_path))) # load last model
print("loading previous model")
criterion = BCEDiceLoss(eps=1.0)
lr = 3e-3
optimiser = torch.optim.SGD(model.parameters(), lr=lr)
scheduler = torch.optim.lr_scheduler.CosineAnnealingWarmRestarts(
optimiser, 7, eta_min=lr / 100, last_epoch=-1)
model = train_model(
model,
train_loader,
valid_loader,
criterion,
optimiser,
scheduler,
save_model_path,
)
if save_model_path.exists():
model.load_state_dict(torch.load(
str(save_model_path))) # load best model
model.eval()
class_parameters = threshold_grid_search(model, valid_loader)
for _, (data, target) in zip(range(2), valid_loader):
data = data.to(global_torch_device(), dtype=torch.float)
output, *_ = model(data)
output = torch.sigmoid(output)
output = output[0].cpu().detach().numpy()
image_vis = data[0].cpu().detach().numpy()
mask = target[0].cpu().detach().numpy()
mask = chw_to_hwc(mask)
output = chw_to_hwc(output)
image_vis = float_chw_to_hwc_uint(image_vis)
pr_mask = numpy.zeros(CloudSegmentationDataset.response_shape)
for j in range(len(CloudSegmentationDataset.categories)):
probability_ = output[..., j]
thr, min_size = class_parameters[j][0], class_parameters[j][1]
pr_mask[..., j], _ = threshold_mask(probability_, thr, min_size)
CloudSegmentationDataset.visualise_prediction(
image_vis,
pr_mask,
original_image=image_vis,
original_mask=mask,
raw_image=image_vis,
raw_mask=output,
)
prepare_submission(model, class_parameters, test_loader)