def main(out_dir=None,
data_dir=None,
cv_data_dir=None,
score_dirs=[],
fusion_method='sum',
prune_imu=None,
standardize=None,
decode=None,
plot_predictions=None,
results_file=None,
sweep_param_name=None,
gpu_dev_id=None,
model_params={},
cv_params={},
train_params={},
viz_params={}):
data_dir = os.path.expanduser(data_dir)
out_dir = os.path.expanduser(out_dir)
score_dirs = tuple(map(os.path.expanduser, score_dirs))
if cv_data_dir is not None:
cv_data_dir = os.path.expanduser(cv_data_dir)
logger = utils.setupRootLogger(filename=os.path.join(out_dir, 'log.txt'))
if results_file is None:
results_file = os.path.join(out_dir, 'results.csv')
else:
results_file = os.path.expanduser(results_file)
fig_dir = os.path.join(out_dir, 'figures')
if not os.path.exists(fig_dir):
os.makedirs(fig_dir)
out_data_dir = os.path.join(out_dir, 'data')
if not os.path.exists(out_data_dir):
os.makedirs(out_data_dir)
def saveVariable(var, var_name):
joblib.dump(var, os.path.join(out_data_dir, f'{var_name}.pkl'))
def loadAll(seq_ids, var_name, data_dir):
def loadOne(seq_id):
fn = os.path.join(data_dir, f'trial={seq_id}_{var_name}')
return joblib.load(fn)
return tuple(map(loadOne, seq_ids))
device = torchutils.selectDevice(gpu_dev_id)
# Load data
dir_trial_ids = tuple(
set(utils.getUniqueIds(d, prefix='trial=', to_array=True))
for d in score_dirs)
dir_trial_ids += (set(
utils.getUniqueIds(data_dir, prefix='trial=', to_array=True)), )
trial_ids = np.array(list(sorted(set.intersection(*dir_trial_ids))))
for dir_name, t_ids in zip(score_dirs + (data_dir, ), dir_trial_ids):
logger.info(f"{len(t_ids)} trial ids from {dir_name}:")
logger.info(f" {t_ids}")
logger.info(f"{len(trial_ids)} trials in intersection: {trial_ids}")
assembly_seqs = loadAll(trial_ids, 'assembly-seq.pkl', data_dir)
feature_seqs = tuple(
loadAll(trial_ids, 'data-scores.pkl', d) for d in score_dirs)
feature_seqs = tuple(zip(*feature_seqs))
# Combine feature seqs
include_indices = []
for i, seq_feats in enumerate(feature_seqs):
feat_shapes = tuple(f.shape for f in seq_feats)
include_seq = all(f == feat_shapes[0] for f in feat_shapes)
if include_seq:
include_indices.append(i)
else:
warn_str = (
f'Excluding trial {trial_ids[i]} with mismatched feature shapes: '
f'{feat_shapes}')
logger.warning(warn_str)
trial_ids = trial_ids[include_indices]
assembly_seqs = tuple(assembly_seqs[i] for i in include_indices)
feature_seqs = tuple(feature_seqs[i] for i in include_indices)
feature_seqs = tuple(np.stack(f) for f in feature_seqs)
# Define cross-validation folds
if cv_data_dir is None:
dataset_size = len(trial_ids)
cv_folds = utils.makeDataSplits(dataset_size, **cv_params)
cv_fold_trial_ids = tuple(
tuple(map(lambda x: trial_ids[x], splits)) for splits in cv_folds)
else:
fn = os.path.join(cv_data_dir, 'cv-fold-trial-ids.pkl')
cv_fold_trial_ids = joblib.load(fn)
def getSplit(split_idxs):
split_data = tuple(
tuple(s[i] for i in split_idxs)
for s in (feature_seqs, assembly_seqs, trial_ids))
return split_data
gt_scores = []
all_scores = []
num_keyframes_total = 0
num_rgb_errors_total = 0
num_correctable_errors_total = 0
num_oov_total = 0
num_changed_total = 0
for cv_index, (train_ids, test_ids) in enumerate(cv_fold_trial_ids):
try:
test_idxs = np.array(
[trial_ids.tolist().index(i) for i in test_ids])
include_indices.append(cv_index)
except ValueError:
logger.info(
f" Skipping fold {cv_index}: missing test data {test_ids}")
logger.info(f'CV fold {cv_index + 1}: {len(trial_ids)} total '
f'({len(train_ids)} train, {len(test_ids)} test)')
# TRAIN PHASE
if cv_data_dir is None:
train_idxs = np.array([trial_ids.index(i) for i in train_ids])
train_assembly_seqs = tuple(assembly_seqs[i] for i in train_idxs)
train_assemblies = []
for seq in train_assembly_seqs:
list(
labels.gen_eq_classes(seq,
train_assemblies,
equivalent=None))
model = None
else:
fn = f'cvfold={cv_index}_train-assemblies.pkl'
train_assemblies = joblib.load(os.path.join(cv_data_dir, fn))
train_idxs = [
i for i in range(len(trial_ids)) if i not in test_idxs
]
fn = f'cvfold={cv_index}_model.pkl'
model = joblib.load(os.path.join(cv_data_dir, fn))
train_features, train_assembly_seqs, train_ids = getSplit(train_idxs)
if False:
train_labels = tuple(
np.array(
list(
labels.gen_eq_classes(assembly_seq,
train_assemblies,
equivalent=None)), )
for assembly_seq in train_assembly_seqs)
train_set = torchutils.SequenceDataset(train_features,
train_labels,
seq_ids=train_ids,
device=device)
train_loader = torch.utils.data.DataLoader(train_set,
batch_size=1,
shuffle=True)
train_epoch_log = collections.defaultdict(list)
# val_epoch_log = collections.defaultdict(list)
metric_dict = {
'Avg Loss': metrics.AverageLoss(),
'Accuracy': metrics.Accuracy()
}
criterion = torch.nn.CrossEntropyLoss()
optimizer_ft = torch.optim.Adam(model.parameters(),
lr=1e-3,
betas=(0.9, 0.999),
eps=1e-08,
weight_decay=0,
amsgrad=False)
lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer_ft,
step_size=1,
gamma=1.00)
model = FusionClassifier(num_sources=train_features[0].shape[0])
model, last_model_wts = torchutils.trainModel(
model,
criterion,
optimizer_ft,
lr_scheduler,
train_loader, # val_loader,
device=device,
metrics=metric_dict,
train_epoch_log=train_epoch_log,
# val_epoch_log=val_epoch_log,
**train_params)
torchutils.plotEpochLog(train_epoch_log,
subfig_size=(10, 2.5),
title='Training performance',
fn=os.path.join(
fig_dir,
f'cvfold={cv_index}_train-plot.png'))
test_assemblies = train_assemblies.copy()
for feature_seq, gt_assembly_seq, trial_id in zip(
*getSplit(test_idxs)):
gt_seq = np.array(
list(
labels.gen_eq_classes(gt_assembly_seq,
test_assemblies,
equivalent=None)))
if plot_predictions:
assembly_fig_dir = os.path.join(fig_dir, 'assembly-imgs')
if not os.path.exists(assembly_fig_dir):
os.makedirs(assembly_fig_dir)
for i, assembly in enumerate(test_assemblies):
assembly.draw(assembly_fig_dir, i)
# TEST PHASE
accuracies = []
for feature_seq, gt_assembly_seq, trial_id in zip(
*getSplit(test_idxs)):
gt_seq = np.array(
list(
labels.gen_eq_classes(gt_assembly_seq,
test_assemblies,
equivalent=None)))
num_labels = gt_seq.shape[0]
num_features = feature_seq.shape[-1]
if num_labels != num_features:
err_str = (f"Skipping trial {trial_id}: "
f"{num_labels} labels != {num_features} features")
logger.info(err_str)
continue
# Ignore OOV states in ground-truth
sample_idxs = np.arange(feature_seq.shape[-1])
score_idxs = gt_seq[gt_seq < feature_seq.shape[1]]
sample_idxs = sample_idxs[gt_seq < feature_seq.shape[1]]
gt_scores.append(feature_seq[:, score_idxs, sample_idxs])
all_scores.append(feature_seq.reshape(feature_seq.shape[0], -1))
if fusion_method == 'sum':
score_seq = feature_seq.sum(axis=0)
elif fusion_method == 'rgb_only':
score_seq = feature_seq[1]
elif fusion_method == 'imu_only':
score_seq = feature_seq[0]
else:
raise NotImplementedError()
if not decode:
model = None
if model is None:
pred_seq = score_seq.argmax(axis=0)
elif isinstance(model, torch.nn.Module):
inputs = torch.tensor(feature_seq[None, ...],
dtype=torch.float,
device=device)
outputs = model.forward(inputs)
pred_seq = model.predict(outputs)[0].cpu().numpy()
else:
dummy_samples = np.arange(score_seq.shape[1])
pred_seq, _, _, _ = model.viterbi(dummy_samples,
log_likelihoods=score_seq,
ml_decode=(not decode))
pred_assemblies = [train_assemblies[i] for i in pred_seq]
gt_assemblies = [test_assemblies[i] for i in gt_seq]
acc = metrics.accuracy_upto(pred_assemblies,
gt_assemblies,
equivalence=None)
accuracies.append(acc)
rgb_pred_seq = feature_seq[1].argmax(axis=0)
num_changed = np.sum(rgb_pred_seq != pred_seq)
rgb_is_wrong = rgb_pred_seq != gt_seq
num_rgb_errors = np.sum(rgb_is_wrong)
imu_scores = feature_seq[0]
imu_scores_gt = np.array([
imu_scores[s_idx,
t] if s_idx < imu_scores.shape[0] else -np.inf
for t, s_idx in enumerate(gt_seq)
])
imu_scores_rgb = np.array([
imu_scores[s_idx,
t] if s_idx < imu_scores.shape[0] else -np.inf
for t, s_idx in enumerate(rgb_pred_seq)
])
# imu_scores_gt = imu_scores[gt_seq, range(len(rgb_pred_seq))]
best_imu_scores = imu_scores.max(axis=0)
imu_is_right = imu_scores_gt >= best_imu_scores
rgb_pred_score_is_lower = imu_scores_gt > imu_scores_rgb
is_correctable_error = rgb_is_wrong & imu_is_right & rgb_pred_score_is_lower
num_correctable_errors = np.sum(is_correctable_error)
prop_correctable = num_correctable_errors / num_rgb_errors
num_oov = np.sum(gt_seq >= len(train_assemblies))
num_states = len(gt_seq)
num_keyframes_total += num_states
num_rgb_errors_total += num_rgb_errors
num_correctable_errors_total += num_correctable_errors
num_oov_total += num_oov
num_changed_total += num_changed
logger.info(f" trial {trial_id}: {num_states} keyframes")
logger.info(f" accuracy (fused): {acc * 100:.1f}%")
logger.info(
f" {num_oov} OOV states ({num_oov / num_states * 100:.1f}%)"
)
logger.info(
f" {num_rgb_errors} RGB errors; "
f"{num_correctable_errors} correctable from IMU ({prop_correctable * 100:.1f}%)"
)
saveVariable(score_seq, f'trial={trial_id}_data-scores')
saveVariable(pred_assemblies,
f'trial={trial_id}_pred-assembly-seq')
saveVariable(gt_assemblies, f'trial={trial_id}_gt-assembly-seq')
if plot_predictions:
io_figs_dir = os.path.join(fig_dir, 'system-io')
if not os.path.exists(io_figs_dir):
os.makedirs(io_figs_dir)
fn = os.path.join(io_figs_dir, f'trial={trial_id:03}.png')
utils.plot_array(feature_seq, (gt_seq, pred_seq, score_seq),
('gt', 'pred', 'scores'),
fn=fn)
score_figs_dir = os.path.join(fig_dir, 'modality-scores')
if not os.path.exists(score_figs_dir):
os.makedirs(score_figs_dir)
plot_scores(feature_seq,
k=25,
fn=os.path.join(score_figs_dir,
f"trial={trial_id:03}.png"))
paths_dir = os.path.join(fig_dir, 'path-imgs')
if not os.path.exists(paths_dir):
os.makedirs(paths_dir)
assemblystats.drawPath(pred_seq, trial_id,
f"trial={trial_id}_pred-seq", paths_dir,
assembly_fig_dir)
assemblystats.drawPath(gt_seq, trial_id,
f"trial={trial_id}_gt-seq", paths_dir,
assembly_fig_dir)
label_seqs = (gt_seq, ) + tuple(
scores.argmax(axis=0) for scores in feature_seq)
label_seqs = np.row_stack(label_seqs)
k = 10
for i, scores in enumerate(feature_seq):
label_score_seqs = tuple(
np.array([
scores[s_idx,
t] if s_idx < scores.shape[0] else -np.inf
for t, s_idx in enumerate(label_seq)
]) for label_seq in label_seqs)
label_score_seqs = np.row_stack(label_score_seqs)
drawPaths(label_seqs,
f"trial={trial_id}_pred-scores_modality={i}",
paths_dir,
assembly_fig_dir,
path_scores=label_score_seqs)
topk_seq = (-scores).argsort(axis=0)[:k, :]
path_scores = np.column_stack(
tuple(scores[idxs, i]
for i, idxs in enumerate(topk_seq.T)))
drawPaths(topk_seq,
f"trial={trial_id}_topk_modality={i}",
paths_dir,
assembly_fig_dir,
path_scores=path_scores)
label_score_seqs = tuple(
np.array([
score_seq[s_idx,
t] if s_idx < score_seq.shape[0] else -np.inf
for t, s_idx in enumerate(label_seq)
]) for label_seq in label_seqs)
label_score_seqs = np.row_stack(label_score_seqs)
drawPaths(label_seqs,
f"trial={trial_id}_pred-scores_fused",
paths_dir,
assembly_fig_dir,
path_scores=label_score_seqs)
topk_seq = (-score_seq).argsort(axis=0)[:k, :]
path_scores = np.column_stack(
tuple(score_seq[idxs, i]
for i, idxs in enumerate(topk_seq.T)))
drawPaths(topk_seq,
f"trial={trial_id}_topk_fused",
paths_dir,
assembly_fig_dir,
path_scores=path_scores)
if accuracies:
fold_accuracy = float(np.array(accuracies).mean())
# logger.info(f' acc: {fold_accuracy * 100:.1f}%')
metric_dict = {'Accuracy': fold_accuracy}
utils.writeResults(results_file, metric_dict, sweep_param_name,
model_params)
num_unexplained_errors = num_rgb_errors_total - (
num_oov_total + num_correctable_errors_total)
prop_correctable = num_correctable_errors_total / num_rgb_errors_total
prop_oov = num_oov_total / num_rgb_errors_total
prop_unexplained = num_unexplained_errors / num_rgb_errors_total
prop_changed = num_changed_total / num_keyframes_total
logger.info("PERFORMANCE ANALYSIS")
logger.info(
f" {num_rgb_errors_total} / {num_keyframes_total} "
f"RGB errors ({num_rgb_errors_total / num_keyframes_total * 100:.1f}%)"
)
logger.info(f" {num_oov_total} / {num_rgb_errors_total} "
f"RGB errors are OOV ({prop_oov * 100:.1f}%)")
logger.info(f" {num_correctable_errors_total} / {num_rgb_errors_total} "
f"RGB errors are correctable ({prop_correctable * 100:.1f}%)")
logger.info(f" {num_unexplained_errors} / {num_rgb_errors_total} "
f"RGB errors are unexplained ({prop_unexplained * 100:.1f}%)")
logger.info(
f" {num_changed_total} / {num_keyframes_total} "
f"Predictions changed after fusion ({prop_changed * 100:.1f}%)")
gt_scores = np.hstack(tuple(gt_scores))
plot_hists(np.exp(gt_scores),
fn=os.path.join(fig_dir, "score-hists_gt.png"))
all_scores = np.hstack(tuple(all_scores))
plot_hists(np.exp(all_scores),
fn=os.path.join(fig_dir, "score-hists_all.png"))
def main(out_dir=None,
data_dir=None,
segs_dir=None,
scores_dir=None,
vocab_dir=None,
label_type='edges',
gpu_dev_id=None,
start_from=None,
stop_at=None,
num_disp_imgs=None,
results_file=None,
sweep_param_name=None,
model_params={},
cv_params={}):
data_dir = os.path.expanduser(data_dir)
segs_dir = os.path.expanduser(segs_dir)
scores_dir = os.path.expanduser(scores_dir)
vocab_dir = os.path.expanduser(vocab_dir)
out_dir = os.path.expanduser(out_dir)
if not os.path.exists(out_dir):
os.makedirs(out_dir)
logger = utils.setupRootLogger(filename=os.path.join(out_dir, 'log.txt'))
if results_file is None:
results_file = os.path.join(out_dir, 'results.csv')
else:
results_file = os.path.expanduser(results_file)
fig_dir = os.path.join(out_dir, 'figures')
if not os.path.exists(fig_dir):
os.makedirs(fig_dir)
io_dir_images = os.path.join(fig_dir, 'model-io_images')
if not os.path.exists(io_dir_images):
os.makedirs(io_dir_images)
io_dir_plots = os.path.join(fig_dir, 'model-io_plots')
if not os.path.exists(io_dir_plots):
os.makedirs(io_dir_plots)
out_data_dir = os.path.join(out_dir, 'data')
if not os.path.exists(out_data_dir):
os.makedirs(out_data_dir)
seq_ids = utils.getUniqueIds(scores_dir,
prefix='trial=',
suffix='score-seq.*',
to_array=True)
logger.info(
f"Loaded scores for {len(seq_ids)} sequences from {scores_dir}")
link_vocab = {}
joint_vocab = {}
joint_type_vocab = {}
vocab, parts_vocab, part_labels = load_vocab(link_vocab, joint_vocab,
joint_type_vocab, vocab_dir)
pred_vocab = [] # FIXME
if label_type == 'assembly':
logger.info("Converting assemblies -> edges")
state_pred_seqs = tuple(
utils.loadVariable(f"trial={seq_id}_pred-label-seq", scores_dir)
for seq_id in seq_ids)
state_true_seqs = tuple(
utils.loadVariable(f"trial={seq_id}_true-label-seq", scores_dir)
for seq_id in seq_ids)
edge_pred_seqs = tuple(part_labels[seq] for seq in state_pred_seqs)
edge_true_seqs = tuple(part_labels[seq] for seq in state_true_seqs)
elif label_type == 'edge':
logger.info("Converting edges -> assemblies (will take a few minutes)")
edge_pred_seqs = tuple(
utils.loadVariable(f"trial={seq_id}_pred-label-seq", scores_dir)
for seq_id in seq_ids)
edge_true_seqs = tuple(
utils.loadVariable(f"trial={seq_id}_true-label-seq", scores_dir)
for seq_id in seq_ids)
state_pred_seqs = tuple(
edges_to_assemblies(seq, pred_vocab, parts_vocab, part_labels)
for seq in edge_pred_seqs)
state_true_seqs = tuple(
edges_to_assemblies(seq, vocab, parts_vocab, part_labels)
for seq in edge_true_seqs)
device = torchutils.selectDevice(gpu_dev_id)
dataset = sim2real.LabeledConnectionDataset(
utils.loadVariable('parts-vocab', vocab_dir),
utils.loadVariable('part-labels', vocab_dir),
utils.loadVariable('vocab', vocab_dir),
device=device)
all_metrics = collections.defaultdict(list)
# Define cross-validation folds
cv_folds = utils.makeDataSplits(len(seq_ids), **cv_params)
utils.saveVariable(cv_folds, 'cv-folds', out_data_dir)
for cv_index, cv_fold in enumerate(cv_folds):
train_indices, val_indices, test_indices = cv_fold
logger.info(
f"CV FOLD {cv_index + 1} / {len(cv_folds)}: "
f"{len(train_indices)} train, {len(val_indices)} val, {len(test_indices)} test"
)
train_states = np.hstack(
tuple(state_true_seqs[i] for i in (train_indices)))
train_edges = part_labels[train_states]
# state_train_vocab = np.unique(train_states)
# edge_train_vocab = part_labels[state_train_vocab]
train_freq_bigram, train_freq_unigram = edge_joint_freqs(train_edges)
# state_probs = utils.makeHistogram(len(vocab), train_states, normalize=True)
test_states = np.hstack(
tuple(state_true_seqs[i] for i in (test_indices)))
test_edges = part_labels[test_states]
# state_test_vocab = np.unique(test_states)
# edge_test_vocab = part_labels[state_test_vocab]
test_freq_bigram, test_freq_unigram = edge_joint_freqs(test_edges)
f, axes = plt.subplots(1, 2)
axes[0].matshow(train_freq_bigram)
axes[0].set_title('Train')
axes[1].matshow(test_freq_bigram)
axes[1].set_title('Test')
plt.tight_layout()
plt.savefig(
os.path.join(fig_dir, f"edge-freqs-bigram_cvfold={cv_index}.png"))
f, axis = plt.subplots(1)
axis.stem(train_freq_unigram,
label='Train',
linefmt='C0-',
markerfmt='C0o')
axis.stem(test_freq_unigram,
label='Test',
linefmt='C1--',
markerfmt='C1o')
plt.legend()
plt.tight_layout()
plt.savefig(
os.path.join(fig_dir, f"edge-freqs-unigram_cvfold={cv_index}.png"))
for i in test_indices:
seq_id = seq_ids[i]
logger.info(f" Processing sequence {seq_id}...")
trial_prefix = f"trial={seq_id}"
# I include the '.' to differentiate between 'rgb-frame-seq' and
# 'rgb-frame-seq-before-first-touch'
# rgb_seq = utils.loadVariable(f"{trial_prefix}_rgb-frame-seq.", data_dir)
# seg_seq = utils.loadVariable(f"{trial_prefix}_seg-labels-seq", segs_dir)
score_seq = utils.loadVariable(f"{trial_prefix}_score-seq",
scores_dir)
# if score_seq.shape[0] != rgb_seq.shape[0]:
# err_str = f"scores shape {score_seq.shape} != data shape {rgb_seq.shape}"
# raise AssertionError(err_str)
edge_pred_seq = edge_pred_seqs[i]
edge_true_seq = edge_true_seqs[i]
state_pred_seq = state_pred_seqs[i]
state_true_seq = state_true_seqs[i]
num_types = np.unique(state_pred_seq).shape[0]
num_samples = state_pred_seq.shape[0]
num_total = len(pred_vocab)
logger.info(
f" {num_types} assemblies predicted ({num_total} total); "
f"{num_samples} samples")
# edge_freq_bigram, edge_freq_unigram = edge_joint_freqs(edge_true_seq)
# dist_shift = np.linalg.norm(train_freq_unigram - edge_freq_unigram)
metric_dict = {
# 'State OOV rate': oov_rate_state(state_true_seq, state_train_vocab),
# 'Edge OOV rate': oov_rate_edges(edge_true_seq, edge_train_vocab),
# 'State avg prob, true': state_probs[state_true_seq].mean(),
# 'State avg prob, pred': state_probs[state_pred_seq].mean(),
# 'Edge distribution shift': dist_shift
}
metric_dict = eval_edge_metrics(edge_pred_seq,
edge_true_seq,
append_to=metric_dict)
metric_dict = eval_state_metrics(state_pred_seq,
state_true_seq,
append_to=metric_dict)
for name, value in metric_dict.items():
logger.info(f" {name}: {value * 100:.2f}%")
all_metrics[name].append(value)
utils.writeResults(results_file, metric_dict, sweep_param_name,
model_params)
if num_disp_imgs is not None:
pred_images = tuple(
render(dataset, vocab[seg_label])
for seg_label in utils.computeSegments(state_pred_seq)[0])
imageprocessing.displayImages(
*pred_images,
file_path=os.path.join(
io_dir_images,
f"seq={seq_id:03d}_pred-assemblies.png"),
num_rows=None,
num_cols=5)
true_images = tuple(
render(dataset, vocab[seg_label])
for seg_label in utils.computeSegments(state_true_seq)[0])
imageprocessing.displayImages(
*true_images,
file_path=os.path.join(
io_dir_images,
f"seq={seq_id:03d}_true-assemblies.png"),
num_rows=None,
num_cols=5)
utils.plot_array(score_seq.T,
(edge_true_seq.T, edge_pred_seq.T),
('true', 'pred'),
fn=os.path.join(io_dir_plots,
f"seq={seq_id:03d}.png"))
def main(out_dir=None,
data_dir=None,
attr_dir=None,
model_name=None,
gpu_dev_id=None,
batch_size=None,
learning_rate=None,
model_params={},
cv_params={},
train_params={},
viz_params={},
plot_predictions=None,
results_file=None,
sweep_param_name=None):
data_dir = os.path.expanduser(data_dir)
out_dir = os.path.expanduser(out_dir)
attr_dir = os.path.expanduser(attr_dir)
if not os.path.exists(out_dir):
os.makedirs(out_dir)
logger = utils.setupRootLogger(filename=os.path.join(out_dir, 'log.txt'))
if results_file is None:
results_file = os.path.join(out_dir, f'results.csv')
else:
results_file = os.path.expanduser(results_file)
fig_dir = os.path.join(out_dir, 'figures')
if not os.path.exists(fig_dir):
os.makedirs(fig_dir)
out_data_dir = os.path.join(out_dir, 'data')
if not os.path.exists(out_data_dir):
os.makedirs(out_data_dir)
def loadData(seq_id):
var_name = f"trial-{seq_id}_rgb-frame-seq"
data = joblib.load(os.path.join(data_dir, f'{var_name}.pkl'))
return data.swapaxes(1, 3)
def loadLabels(seq_id):
var_name = f"trial-{seq_id}_label-seq"
return joblib.load(os.path.join(attr_dir, f'{var_name}.pkl'))
def saveVariable(var, var_name):
joblib.dump(var, os.path.join(out_data_dir, f'{var_name}.pkl'))
# Load data
trial_ids = utils.getUniqueIds(data_dir)
label_seqs = tuple(map(loadLabels, trial_ids))
device = torchutils.selectDevice(gpu_dev_id)
# Define cross-validation folds
dataset_size = len(trial_ids)
cv_folds = utils.makeDataSplits(dataset_size, **cv_params)
def getSplit(split_idxs):
split_data = tuple(
tuple(s[i] for i in split_idxs) for s in (label_seqs, trial_ids))
return split_data
for cv_index, cv_splits in enumerate(cv_folds):
train_data, val_data, test_data = tuple(map(getSplit, cv_splits))
criterion = torch.nn.BCEWithLogitsLoss()
labels_dtype = torch.float
train_labels, train_ids = train_data
train_set = torchutils.PickledVideoDataset(loadData,
train_labels,
device=device,
labels_dtype=labels_dtype,
seq_ids=train_ids,
batch_size=batch_size)
train_loader = torch.utils.data.DataLoader(train_set,
batch_size=1,
shuffle=True)
test_labels, test_ids = test_data
test_set = torchutils.PickledVideoDataset(loadData,
test_labels,
device=device,
labels_dtype=labels_dtype,
seq_ids=test_ids,
batch_size=batch_size)
test_loader = torch.utils.data.DataLoader(test_set,
batch_size=1,
shuffle=False)
val_labels, val_ids = val_data
val_set = torchutils.PickledVideoDataset(loadData,
val_labels,
device=device,
labels_dtype=labels_dtype,
seq_ids=val_ids,
batch_size=batch_size)
val_loader = torch.utils.data.DataLoader(val_set,
batch_size=1,
shuffle=True)
logger.info(
f'CV fold {cv_index + 1} / {len(cv_folds)}: {len(trial_ids)} total '
f'({len(train_ids)} train, {len(val_ids)} val, {len(test_ids)} test)'
)
if model_name == 'resnet':
# input_dim = train_set.num_obsv_dims
output_dim = train_set.num_label_types
model = ImageClassifier(output_dim,
**model_params).to(device=device)
else:
raise AssertionError()
train_epoch_log = collections.defaultdict(list)
val_epoch_log = collections.defaultdict(list)
metric_dict = {
'Avg Loss': metrics.AverageLoss(),
'Accuracy': metrics.Accuracy(),
'Precision': metrics.Precision(),
'Recall': metrics.Recall(),
'F1': metrics.Fmeasure()
}
optimizer_ft = torch.optim.Adam(model.parameters(),
lr=learning_rate,
betas=(0.9, 0.999),
eps=1e-08,
weight_decay=0,
amsgrad=False)
lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer_ft,
step_size=1,
gamma=1.00)
model, last_model_wts = torchutils.trainModel(
model,
criterion,
optimizer_ft,
lr_scheduler,
train_loader,
val_loader,
device=device,
metrics=metric_dict,
train_epoch_log=train_epoch_log,
val_epoch_log=val_epoch_log,
**train_params)
# Test model
metric_dict = {
'Avg Loss': metrics.AverageLoss(),
'Accuracy': metrics.Accuracy(),
'Precision': metrics.Precision(),
'Recall': metrics.Recall(),
'F1': metrics.Fmeasure()
}
test_io_history = torchutils.predictSamples(
model.to(device=device),
test_loader,
criterion=criterion,
device=device,
metrics=metric_dict,
data_labeled=True,
update_model=False,
seq_as_batch=train_params['seq_as_batch'],
return_io_history=True)
metric_str = ' '.join(str(m) for m in metric_dict.values())
logger.info('[TST] ' + metric_str)
utils.writeResults(results_file, metric_dict, sweep_param_name,
model_params)
if plot_predictions:
# imu.plot_prediction_eg(test_io_history, fig_dir, fig_type=fig_type, **viz_params)
imu.plot_prediction_eg(test_io_history, fig_dir, **viz_params)
def saveTrialData(pred_seq, score_seq, feat_seq, label_seq, trial_id):
saveVariable(pred_seq.cpu().numpy(),
f'trial={trial_id}_pred-label-seq')
saveVariable(score_seq.cpu().numpy(),
f'trial={trial_id}_score-seq')
saveVariable(label_seq.cpu().numpy(),
f'trial={trial_id}_true-label-seq')
for io in test_io_history:
saveTrialData(*io)
saveVariable(train_ids, f'cvfold={cv_index}_train-ids')
saveVariable(test_ids, f'cvfold={cv_index}_test-ids')
saveVariable(val_ids, f'cvfold={cv_index}_val-ids')
saveVariable(train_epoch_log,
f'cvfold={cv_index}_{model_name}-train-epoch-log')
saveVariable(val_epoch_log,
f'cvfold={cv_index}_{model_name}-val-epoch-log')
saveVariable(metric_dict,
f'cvfold={cv_index}_{model_name}-metric-dict')
saveVariable(model, f'cvfold={cv_index}_{model_name}-best')
model.load_state_dict(last_model_wts)
saveVariable(model, f'cvfold={cv_index}_{model_name}-last')
torchutils.plotEpochLog(train_epoch_log,
subfig_size=(10, 2.5),
title='Training performance',
fn=os.path.join(
fig_dir,
f'cvfold={cv_index}_train-plot.png'))
if val_epoch_log:
torchutils.plotEpochLog(val_epoch_log,
subfig_size=(10, 2.5),
title='Heldout performance',
fn=os.path.join(
fig_dir,
f'cvfold={cv_index}_val-plot.png'))
def main(out_dir=None,
data_dir=None,
model_name=None,
gpu_dev_id=None,
batch_size=None,
learning_rate=None,
model_params={},
cv_params={},
train_params={},
viz_params={},
load_masks_params={},
kornia_tfs={},
only_edge=None,
num_disp_imgs=None,
results_file=None,
sweep_param_name=None):
data_dir = os.path.expanduser(data_dir)
out_dir = os.path.expanduser(out_dir)
if not os.path.exists(out_dir):
os.makedirs(out_dir)
logger = utils.setupRootLogger(filename=os.path.join(out_dir, 'log.txt'))
if results_file is None:
results_file = os.path.join(out_dir, 'results.csv')
else:
results_file = os.path.expanduser(results_file)
fig_dir = os.path.join(out_dir, 'figures')
if not os.path.exists(fig_dir):
os.makedirs(fig_dir)
io_dir = os.path.join(fig_dir, 'model-io')
if not os.path.exists(io_dir):
os.makedirs(io_dir)
out_data_dir = os.path.join(out_dir, 'data')
if not os.path.exists(out_data_dir):
os.makedirs(out_data_dir)
def saveVariable(var, var_name, to_dir=out_data_dir):
return utils.saveVariable(var, var_name, to_dir)
trial_ids = utils.getUniqueIds(data_dir, prefix='trial=', to_array=True)
vocab = [BlockAssembly()
] + [make_single_block_state(i) for i in range(len(defn.blocks))]
for seq_id in trial_ids:
assembly_seq = utils.loadVariable(f"trial={seq_id}_assembly-seq",
data_dir)
for assembly in assembly_seq:
utils.getIndex(assembly, vocab)
parts_vocab, part_labels = labels_lib.make_parts_vocab(
vocab, lower_tri_only=True, append_to_vocab=True)
if only_edge is not None:
part_labels = part_labels[:, only_edge:only_edge + 1]
logger.info(
f"Loaded {len(trial_ids)} sequences; {len(vocab)} unique assemblies")
saveVariable(vocab, 'vocab')
saveVariable(parts_vocab, 'parts-vocab')
saveVariable(part_labels, 'part-labels')
device = torchutils.selectDevice(gpu_dev_id)
if model_name == 'AAE':
Dataset = sim2real.DenoisingDataset
elif model_name == 'Resnet':
Dataset = sim2real.RenderDataset
elif model_name == 'Connections':
Dataset = sim2real.ConnectionDataset
elif model_name == 'Labeled Connections':
Dataset = sim2real.LabeledConnectionDataset
occlusion_masks = loadMasks(**load_masks_params)
if occlusion_masks is not None:
logger.info(f"Loaded {occlusion_masks.shape[0]} occlusion masks")
def make_data(shuffle=True):
dataset = Dataset(
parts_vocab,
part_labels,
vocab,
device=device,
occlusion_masks=occlusion_masks,
kornia_tfs=kornia_tfs,
)
data_loader = torch.utils.data.DataLoader(dataset,
batch_size=batch_size,
shuffle=shuffle)
return dataset, data_loader
for cv_index, cv_splits in enumerate(range(1)):
cv_str = f"cvfold={cv_index}"
train_set, train_loader = make_data(shuffle=True)
test_set, test_loader = make_data(shuffle=False)
val_set, val_loader = make_data(shuffle=True)
if model_name == 'AAE':
model = sim2real.AugmentedAutoEncoder(train_set.data_shape,
train_set.num_classes)
criterion = torchutils.BootstrappedCriterion(
0.25,
base_criterion=torch.nn.functional.mse_loss,
)
metric_names = ('Reciprocal Loss', )
elif model_name == 'Resnet':
model = sim2real.ImageClassifier(train_set.num_classes,
**model_params)
criterion = torch.nn.CrossEntropyLoss()
metric_names = ('Loss', 'Accuracy')
elif model_name == 'Connections':
model = sim2real.ConnectionClassifier(train_set.label_shape[0],
**model_params)
criterion = torch.nn.BCEWithLogitsLoss()
metric_names = ('Loss', 'Accuracy', 'Precision', 'Recall', 'F1')
elif model_name == 'Labeled Connections':
out_dim = int(part_labels.max()) + 1
num_vertices = len(defn.blocks)
edges = np.column_stack(np.tril_indices(num_vertices, k=-1))
if only_edge is not None:
edges = edges[only_edge:only_edge + 1]
model = sim2real.LabeledConnectionClassifier(
out_dim, num_vertices, edges, **model_params)
if only_edge is not None:
logger.info(f"Class freqs: {train_set.class_freqs}")
# criterion = torch.nn.CrossEntropyLoss(weight=1 / train_set.class_freqs[:, 0])
criterion = torch.nn.CrossEntropyLoss()
else:
criterion = torch.nn.CrossEntropyLoss()
# criterion = torchutils.BootstrappedCriterion(
# 0.25, base_criterion=torch.nn.functional.cross_entropy,
# )
metric_names = ('Loss', 'Accuracy', 'Precision', 'Recall', 'F1')
model = model.to(device=device)
optimizer_ft = torch.optim.Adam(model.parameters(),
lr=learning_rate,
betas=(0.9, 0.999),
eps=1e-08,
weight_decay=0,
amsgrad=False)
lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer_ft,
step_size=1,
gamma=1.00)
train_epoch_log = collections.defaultdict(list)
val_epoch_log = collections.defaultdict(list)
metric_dict = {name: metrics.makeMetric(name) for name in metric_names}
model, last_model_wts = torchutils.trainModel(
model,
criterion,
optimizer_ft,
lr_scheduler,
train_loader,
val_loader,
device=device,
metrics=metric_dict,
train_epoch_log=train_epoch_log,
val_epoch_log=val_epoch_log,
**train_params)
# Test model
metric_dict = {name: metrics.makeMetric(name) for name in metric_names}
test_io_batches = torchutils.predictSamples(
model.to(device=device),
test_loader,
criterion=criterion,
device=device,
metrics=metric_dict,
data_labeled=True,
update_model=False,
seq_as_batch=train_params['seq_as_batch'],
return_io_history=True)
metric_str = ' '.join(str(m) for m in metric_dict.values())
logger.info('[TST] ' + metric_str)
utils.writeResults(results_file, metric_dict, sweep_param_name,
model_params)
for pred_seq, score_seq, feat_seq, label_seq, trial_id in test_io_batches:
trial_str = f"trial={trial_id}"
saveVariable(pred_seq.cpu().numpy(), f'{trial_str}_pred-label-seq')
saveVariable(score_seq.cpu().numpy(), f'{trial_str}_score-seq')
saveVariable(label_seq.cpu().numpy(),
f'{trial_str}_true-label-seq')
saveVariable(model, f'{cv_str}_model-best')
if train_epoch_log:
torchutils.plotEpochLog(train_epoch_log,
subfig_size=(10, 2.5),
title='Training performance',
fn=os.path.join(
fig_dir, f'{cv_str}_train-plot.png'))
if val_epoch_log:
torchutils.plotEpochLog(val_epoch_log,
subfig_size=(10, 2.5),
title='Heldout performance',
fn=os.path.join(fig_dir,
f'{cv_str}_val-plot.png'))
if num_disp_imgs is not None:
model.plotBatches(test_io_batches, io_dir, dataset=test_set)
def main(out_dir=None,
rgb_data_dir=None,
rgb_attributes_dir=None,
rgb_vocab_dir=None,
imu_data_dir=None,
imu_attributes_dir=None,
modalities=['rgb', 'imu'],
gpu_dev_id=None,
plot_predictions=None,
results_file=None,
sweep_param_name=None,
model_params={},
cv_params={},
train_params={},
viz_params={}):
out_dir = os.path.expanduser(out_dir)
rgb_data_dir = os.path.expanduser(rgb_data_dir)
rgb_attributes_dir = os.path.expanduser(rgb_attributes_dir)
rgb_vocab_dir = os.path.expanduser(rgb_vocab_dir)
imu_data_dir = os.path.expanduser(imu_data_dir)
imu_attributes_dir = os.path.expanduser(imu_attributes_dir)
logger = utils.setupRootLogger(filename=os.path.join(out_dir, 'log.txt'))
if results_file is None:
results_file = os.path.join(out_dir, 'results.csv')
else:
results_file = os.path.expanduser(results_file)
fig_dir = os.path.join(out_dir, 'figures')
if not os.path.exists(fig_dir):
os.makedirs(fig_dir)
out_data_dir = os.path.join(out_dir, 'data')
if not os.path.exists(out_data_dir):
os.makedirs(out_data_dir)
def saveVariable(var, var_name, to_dir=out_data_dir):
utils.saveVariable(var, var_name, to_dir)
# Load data
if modalities == ['rgb']:
trial_ids = utils.getUniqueIds(rgb_data_dir,
prefix='trial=',
to_array=True)
logger.info(f"Processing {len(trial_ids)} videos")
else:
rgb_trial_ids = utils.getUniqueIds(rgb_data_dir,
prefix='trial=',
to_array=True)
imu_trial_ids = utils.getUniqueIds(imu_data_dir,
prefix='trial=',
to_array=True)
trial_ids = np.array(
sorted(set(rgb_trial_ids.tolist()) & set(imu_trial_ids.tolist())))
logger.info(
f"Processing {len(trial_ids)} videos common to "
f"RGB ({len(rgb_trial_ids)} total) and IMU ({len(imu_trial_ids)} total)"
)
device = torchutils.selectDevice(gpu_dev_id)
dataset = FusionDataset(trial_ids,
rgb_attributes_dir,
rgb_data_dir,
imu_attributes_dir,
imu_data_dir,
device=device,
modalities=modalities)
utils.saveMetadata(dataset.metadata, out_data_dir)
saveVariable(dataset.vocab, 'vocab')
# parts_vocab = loadVariable('parts-vocab')
edge_labels = {
'rgb':
utils.loadVariable('part-labels', rgb_vocab_dir),
'imu':
np.stack([
labels.inSameComponent(a, lower_tri_only=True)
for a in dataset.vocab
])
}
# edge_labels = revise_edge_labels(edge_labels, input_seqs)
attribute_labels = tuple(edge_labels[name] for name in modalities)
logger.info('Making transition probs...')
transition_probs = make_transition_scores(dataset.vocab)
saveVariable(transition_probs, 'transition-probs')
model = AttributeModel(*attribute_labels, device=device)
if plot_predictions:
figsize = (12, 3)
fig, axis = plt.subplots(1, figsize=figsize)
axis.imshow(edge_labels['rgb'].T, interpolation='none', aspect='auto')
plt.savefig(os.path.join(fig_dir, "edge-labels.png"))
plt.close()
for i, trial_id in enumerate(trial_ids):
logger.info(f"Processing sequence {trial_id}...")
trial_prefix = f"trial={trial_id}"
true_label_seq = dataset.loadTargets(trial_id)
attribute_feats = dataset.loadInputs(trial_id)
score_seq = model(attribute_feats)
pred_label_seq = model.predict(score_seq)
attribute_feats = attribute_feats.cpu().numpy()
score_seq = score_seq.cpu().numpy()
true_label_seq = true_label_seq.cpu().numpy()
pred_label_seq = pred_label_seq.cpu().numpy()
saveVariable(score_seq.T, f'{trial_prefix}_score-seq')
saveVariable(true_label_seq.T, f'{trial_prefix}_label-seq')
if plot_predictions:
fn = os.path.join(fig_dir, f'{trial_prefix}.png')
utils.plot_array(attribute_feats.T,
(true_label_seq, pred_label_seq, score_seq),
('gt', 'pred', 'scores'),
fn=fn)
metric_dict = eval_metrics(pred_label_seq, true_label_seq)
for name, value in metric_dict.items():
logger.info(f" {name}: {value * 100:.2f}%")
utils.writeResults(results_file, metric_dict, sweep_param_name,
model_params)
def main(out_dir=None,
data_dir=None,
model_name=None,
predict_mode='classify',
gpu_dev_id=None,
batch_size=None,
learning_rate=None,
independent_signals=None,
active_only=None,
output_dim_from_vocab=False,
prefix='trial=',
feature_fn_format='feature-seq.pkl',
label_fn_format='label_seq.pkl',
dataset_params={},
model_params={},
cv_params={},
train_params={},
viz_params={},
metric_names=['Loss', 'Accuracy', 'Precision', 'Recall', 'F1'],
plot_predictions=None,
results_file=None,
sweep_param_name=None):
data_dir = os.path.expanduser(data_dir)
out_dir = os.path.expanduser(out_dir)
if not os.path.exists(out_dir):
os.makedirs(out_dir)
logger = utils.setupRootLogger(filename=os.path.join(out_dir, 'log.txt'))
if results_file is None:
results_file = os.path.join(out_dir, 'results.csv')
else:
results_file = os.path.expanduser(results_file)
fig_dir = os.path.join(out_dir, 'figures')
if not os.path.exists(fig_dir):
os.makedirs(fig_dir)
out_data_dir = os.path.join(out_dir, 'data')
if not os.path.exists(out_data_dir):
os.makedirs(out_data_dir)
def saveVariable(var, var_name, to_dir=out_data_dir):
return utils.saveVariable(var, var_name, to_dir)
# Load data
device = torchutils.selectDevice(gpu_dev_id)
trial_ids = utils.getUniqueIds(data_dir,
prefix=prefix,
suffix=feature_fn_format,
to_array=True)
dataset = utils.CvDataset(
trial_ids,
data_dir,
prefix=prefix,
feature_fn_format=feature_fn_format,
label_fn_format=label_fn_format,
)
utils.saveMetadata(dataset.metadata, out_data_dir)
utils.saveVariable(dataset.vocab, 'vocab', out_data_dir)
# Define cross-validation folds
cv_folds = utils.makeDataSplits(len(trial_ids), **cv_params)
utils.saveVariable(cv_folds, 'cv-folds', out_data_dir)
if predict_mode == 'binary multiclass':
# criterion = torchutils.BootstrappedCriterion(
# 0.25, base_criterion=torch.nn.functional.binary_cross_entropy_with_logits,
# )
criterion = torch.nn.BCEWithLogitsLoss()
labels_dtype = torch.float
elif predict_mode == 'multiclass':
criterion = torch.nn.CrossEntropyLoss()
labels_dtype = torch.long
elif predict_mode == 'classify':
criterion = torch.nn.CrossEntropyLoss()
labels_dtype = torch.long
else:
raise AssertionError()
def make_dataset(feats, labels, ids, shuffle=True):
dataset = torchutils.SequenceDataset(feats,
labels,
device=device,
labels_dtype=labels_dtype,
seq_ids=ids,
**dataset_params)
loader = torch.utils.data.DataLoader(dataset,
batch_size=batch_size,
shuffle=True)
return dataset, loader
for cv_index, cv_fold in enumerate(cv_folds):
train_data, val_data, test_data = dataset.getFold(cv_fold)
if independent_signals:
train_data = splitSeqs(*train_data, active_only=active_only)
val_data = splitSeqs(*val_data, active_only=active_only)
test_data = splitSeqs(*test_data, active_only=False)
train_set, train_loader = make_dataset(*train_data, shuffle=True)
test_set, test_loader = make_dataset(*test_data, shuffle=False)
val_set, val_loader = make_dataset(*val_data, shuffle=True)
logger.info(
f'CV fold {cv_index + 1} / {len(cv_folds)}: {len(dataset.trial_ids)} total '
f'({len(train_set)} train, {len(val_set)} val, {len(test_set)} test)'
)
logger.info(f'{train_set.num_label_types} unique labels in train set; '
f'vocab size is {len(dataset.vocab)}')
input_dim = train_set.num_obsv_dims
output_dim = train_set.num_label_types
if output_dim_from_vocab:
output_dim = len(dataset.vocab)
if model_name == 'linear':
model = torchutils.LinearClassifier(
input_dim, output_dim, **model_params).to(device=device)
elif model_name == 'conv':
model = ConvClassifier(input_dim, output_dim,
**model_params).to(device=device)
elif model_name == 'TCN':
if predict_mode == 'multiclass':
num_multiclass = train_set[0][1].shape[-1]
output_dim = max([
train_set.num_label_types, test_set.num_label_types,
val_set.num_label_types
])
else:
num_multiclass = None
model = TcnClassifier(input_dim,
output_dim,
num_multiclass=num_multiclass,
**model_params).to(device=device)
elif model_name == 'LSTM':
if predict_mode == 'multiclass':
num_multiclass = train_set[0][1].shape[-1]
output_dim = max([
train_set.num_label_types, test_set.num_label_types,
val_set.num_label_types
])
else:
num_multiclass = None
model = LstmClassifier(input_dim,
output_dim,
num_multiclass=num_multiclass,
**model_params).to(device=device)
else:
raise AssertionError()
optimizer_ft = torch.optim.Adam(model.parameters(),
lr=learning_rate,
betas=(0.9, 0.999),
eps=1e-08,
weight_decay=0,
amsgrad=False)
lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer_ft,
step_size=1,
gamma=1.00)
train_epoch_log = collections.defaultdict(list)
val_epoch_log = collections.defaultdict(list)
metric_dict = {name: metrics.makeMetric(name) for name in metric_names}
model, last_model_wts = torchutils.trainModel(
model,
criterion,
optimizer_ft,
lr_scheduler,
train_loader,
val_loader,
device=device,
metrics=metric_dict,
train_epoch_log=train_epoch_log,
val_epoch_log=val_epoch_log,
**train_params)
# Test model
metric_dict = {name: metrics.makeMetric(name) for name in metric_names}
test_io_history = torchutils.predictSamples(
model.to(device=device),
test_loader,
criterion=criterion,
device=device,
metrics=metric_dict,
data_labeled=True,
update_model=False,
seq_as_batch=train_params['seq_as_batch'],
return_io_history=True)
if independent_signals:
test_io_history = tuple(joinSeqs(test_io_history))
logger.info('[TST] ' +
' '.join(str(m) for m in metric_dict.values()))
utils.writeResults(results_file,
{k: v.value
for k, v in metric_dict.items()}, sweep_param_name,
model_params)
if plot_predictions:
io_fig_dir = os.path.join(fig_dir, 'model-io')
if not os.path.exists(io_fig_dir):
os.makedirs(io_fig_dir)
label_names = ('gt', 'pred')
preds, scores, inputs, gt_labels, ids = zip(*test_io_history)
for batch in test_io_history:
batch = tuple(
x.cpu().numpy() if isinstance(x, torch.Tensor) else x
for x in batch)
for preds, _, inputs, gt_labels, seq_id in zip(*batch):
fn = os.path.join(io_fig_dir,
f"{prefix}{seq_id}_model-io.png")
utils.plot_array(inputs, (gt_labels.T, preds.T),
label_names,
fn=fn)
for batch in test_io_history:
batch = tuple(x.cpu().numpy() if isinstance(x, torch.Tensor) else x
for x in batch)
for pred_seq, score_seq, feat_seq, label_seq, trial_id in zip(
*batch):
saveVariable(pred_seq, f'{prefix}{trial_id}_pred-label-seq')
saveVariable(score_seq, f'{prefix}{trial_id}_score-seq')
saveVariable(label_seq, f'{prefix}{trial_id}_true-label-seq')
saveVariable(model, f'cvfold={cv_index}_{model_name}-best')
train_fig_dir = os.path.join(fig_dir, 'train-plots')
if not os.path.exists(train_fig_dir):
os.makedirs(train_fig_dir)
if train_epoch_log:
torchutils.plotEpochLog(train_epoch_log,
subfig_size=(10, 2.5),
title='Training performance',
fn=os.path.join(
train_fig_dir,
f'cvfold={cv_index}_train-plot.png'))
if val_epoch_log:
torchutils.plotEpochLog(val_epoch_log,
subfig_size=(10, 2.5),
title='Heldout performance',
fn=os.path.join(
train_fig_dir,
f'cvfold={cv_index}_val-plot.png'))
def main(out_dir=None,
data_dir=None,
model_name=None,
pretrained_model_dir=None,
gpu_dev_id=None,
batch_size=None,
learning_rate=None,
independent_signals=None,
active_only=None,
model_params={},
cv_params={},
train_params={},
viz_params={},
plot_predictions=None,
results_file=None,
sweep_param_name=None,
label_mapping=None,
eval_label_mapping=None):
data_dir = os.path.expanduser(data_dir)
out_dir = os.path.expanduser(out_dir)
if not os.path.exists(out_dir):
os.makedirs(out_dir)
logger = utils.setupRootLogger(filename=os.path.join(out_dir, 'log.txt'))
if results_file is None:
results_file = os.path.join(out_dir, 'results.csv')
else:
results_file = os.path.expanduser(results_file)
fig_dir = os.path.join(out_dir, 'figures')
if not os.path.exists(fig_dir):
os.makedirs(fig_dir)
out_data_dir = os.path.join(out_dir, 'data')
if not os.path.exists(out_data_dir):
os.makedirs(out_data_dir)
def saveVariable(var, var_name):
joblib.dump(var, os.path.join(out_data_dir, f'{var_name}.pkl'))
def loadAll(seq_ids, var_name, data_dir):
def loadOne(seq_id):
fn = os.path.join(data_dir, f'trial={seq_id}_{var_name}')
return joblib.load(fn)
return tuple(map(loadOne, seq_ids))
# Load data
trial_ids = utils.getUniqueIds(data_dir, prefix='trial=', to_array=True)
feature_seqs = loadAll(trial_ids, 'feature-seq.pkl', data_dir)
label_seqs = loadAll(trial_ids, 'label-seq.pkl', data_dir)
device = torchutils.selectDevice(gpu_dev_id)
if label_mapping is not None:
def map_labels(labels):
for i, j in label_mapping.items():
labels[labels == i] = j
return labels
label_seqs = tuple(map(map_labels, label_seqs))
# Define cross-validation folds
dataset_size = len(trial_ids)
cv_folds = utils.makeDataSplits(dataset_size, **cv_params)
def getSplit(split_idxs):
split_data = tuple(
tuple(s[i] for i in split_idxs)
for s in (feature_seqs, label_seqs, trial_ids))
return split_data
for cv_index, cv_splits in enumerate(cv_folds):
if pretrained_model_dir is not None:
def loadFromPretrain(fn):
return joblib.load(
os.path.join(pretrained_model_dir, f"{fn}.pkl"))
model = loadFromPretrain(f'cvfold={cv_index}_{model_name}-best')
train_ids = loadFromPretrain(f'cvfold={cv_index}_train-ids')
val_ids = loadFromPretrain(f'cvfold={cv_index}_val-ids')
test_ids = tuple(i for i in trial_ids
if i not in (train_ids + val_ids))
test_idxs = tuple(trial_ids.tolist().index(i) for i in test_ids)
test_data = getSplit(test_idxs)
if independent_signals:
criterion = torch.nn.CrossEntropyLoss()
labels_dtype = torch.long
test_data = splitSeqs(*test_data, active_only=False)
else:
# FIXME
# criterion = torch.nn.BCEWithLogitsLoss()
# labels_dtype = torch.float
criterion = torch.nn.CrossEntropyLoss()
labels_dtype = torch.long
test_feats, test_labels, test_ids = test_data
test_set = torchutils.SequenceDataset(test_feats,
test_labels,
device=device,
labels_dtype=labels_dtype,
seq_ids=test_ids,
transpose_data=True)
test_loader = torch.utils.data.DataLoader(test_set,
batch_size=batch_size,
shuffle=False)
# Test model
metric_dict = {
'Avg Loss': metrics.AverageLoss(),
'Accuracy': metrics.Accuracy(),
'Precision': metrics.Precision(),
'Recall': metrics.Recall(),
'F1': metrics.Fmeasure()
}
test_io_history = torchutils.predictSamples(
model.to(device=device),
test_loader,
criterion=criterion,
device=device,
metrics=metric_dict,
data_labeled=True,
update_model=False,
seq_as_batch=train_params['seq_as_batch'],
return_io_history=True)
if independent_signals:
test_io_history = tuple(joinSeqs(test_io_history))
metric_str = ' '.join(str(m) for m in metric_dict.values())
logger.info('[TST] ' + metric_str)
d = {k: v.value for k, v in metric_dict.items()}
utils.writeResults(results_file, d, sweep_param_name, model_params)
if plot_predictions:
imu.plot_prediction_eg(test_io_history, fig_dir, **viz_params)
def saveTrialData(pred_seq, score_seq, feat_seq, label_seq,
trial_id):
if label_mapping is not None:
def dup_score_cols(scores):
num_cols = scores.shape[-1] + len(label_mapping)
col_idxs = torch.arange(num_cols)
for i, j in label_mapping.items():
col_idxs[i] = j
return scores[..., col_idxs]
score_seq = dup_score_cols(score_seq)
saveVariable(pred_seq.cpu().numpy(),
f'trial={trial_id}_pred-label-seq')
saveVariable(score_seq.cpu().numpy(),
f'trial={trial_id}_score-seq')
saveVariable(label_seq.cpu().numpy(),
f'trial={trial_id}_true-label-seq')
for io in test_io_history:
saveTrialData(*io)
continue
train_data, val_data, test_data = tuple(map(getSplit, cv_splits))
if independent_signals:
criterion = torch.nn.CrossEntropyLoss()
labels_dtype = torch.long
split_ = functools.partial(splitSeqs, active_only=active_only)
train_data = split_(*train_data)
val_data = split_(*val_data)
test_data = splitSeqs(*test_data, active_only=False)
else:
# FIXME
# criterion = torch.nn.BCEWithLogitsLoss()
# labels_dtype = torch.float
criterion = torch.nn.CrossEntropyLoss()
labels_dtype = torch.long
train_feats, train_labels, train_ids = train_data
train_set = torchutils.SequenceDataset(train_feats,
train_labels,
device=device,
labels_dtype=labels_dtype,
seq_ids=train_ids,
transpose_data=True)
train_loader = torch.utils.data.DataLoader(train_set,
batch_size=batch_size,
shuffle=True)
test_feats, test_labels, test_ids = test_data
test_set = torchutils.SequenceDataset(test_feats,
test_labels,
device=device,
labels_dtype=labels_dtype,
seq_ids=test_ids,
transpose_data=True)
test_loader = torch.utils.data.DataLoader(test_set,
batch_size=batch_size,
shuffle=False)
val_feats, val_labels, val_ids = val_data
val_set = torchutils.SequenceDataset(val_feats,
val_labels,
device=device,
labels_dtype=labels_dtype,
seq_ids=val_ids,
transpose_data=True)
val_loader = torch.utils.data.DataLoader(val_set,
batch_size=batch_size,
shuffle=True)
logger.info(
f'CV fold {cv_index + 1} / {len(cv_folds)}: {len(trial_ids)} total '
f'({len(train_ids)} train, {len(val_ids)} val, {len(test_ids)} test)'
)
input_dim = train_set.num_obsv_dims
output_dim = train_set.num_label_types
if model_name == 'linear':
model = torchutils.LinearClassifier(
input_dim, output_dim, **model_params).to(device=device)
elif model_name == 'conv':
model = ConvClassifier(input_dim, output_dim,
**model_params).to(device=device)
elif model_name == 'TCN':
model = TcnClassifier(input_dim, output_dim, **model_params)
else:
raise AssertionError()
train_epoch_log = collections.defaultdict(list)
val_epoch_log = collections.defaultdict(list)
metric_dict = {
'Avg Loss': metrics.AverageLoss(),
'Accuracy': metrics.Accuracy(),
'Precision': metrics.Precision(),
'Recall': metrics.Recall(),
'F1': metrics.Fmeasure()
}
optimizer_ft = torch.optim.Adam(model.parameters(),
lr=learning_rate,
betas=(0.9, 0.999),
eps=1e-08,
weight_decay=0,
amsgrad=False)
lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer_ft,
step_size=1,
gamma=1.00)
model, last_model_wts = torchutils.trainModel(
model,
criterion,
optimizer_ft,
lr_scheduler,
train_loader,
val_loader,
device=device,
metrics=metric_dict,
train_epoch_log=train_epoch_log,
val_epoch_log=val_epoch_log,
**train_params)
# Test model
metric_dict = {
'Avg Loss': metrics.AverageLoss(),
'Accuracy': metrics.Accuracy(),
'Precision': metrics.Precision(),
'Recall': metrics.Recall(),
'F1': metrics.Fmeasure()
}
test_io_history = torchutils.predictSamples(
model.to(device=device),
test_loader,
criterion=criterion,
device=device,
metrics=metric_dict,
data_labeled=True,
update_model=False,
seq_as_batch=train_params['seq_as_batch'],
return_io_history=True)
if independent_signals:
test_io_history = tuple(joinSeqs(test_io_history))
metric_str = ' '.join(str(m) for m in metric_dict.values())
logger.info('[TST] ' + metric_str)
d = {k: v.value for k, v in metric_dict.items()}
utils.writeResults(results_file, d, sweep_param_name, model_params)
if plot_predictions:
# imu.plot_prediction_eg(test_io_history, fig_dir, fig_type=fig_type, **viz_params)
imu.plot_prediction_eg(test_io_history, fig_dir, **viz_params)
def saveTrialData(pred_seq, score_seq, feat_seq, label_seq, trial_id):
if label_mapping is not None:
def dup_score_cols(scores):
num_cols = scores.shape[-1] + len(label_mapping)
col_idxs = torch.arange(num_cols)
for i, j in label_mapping.items():
col_idxs[i] = j
return scores[..., col_idxs]
score_seq = dup_score_cols(score_seq)
saveVariable(pred_seq.cpu().numpy(),
f'trial={trial_id}_pred-label-seq')
saveVariable(score_seq.cpu().numpy(),
f'trial={trial_id}_score-seq')
saveVariable(label_seq.cpu().numpy(),
f'trial={trial_id}_true-label-seq')
for io in test_io_history:
saveTrialData(*io)
saveVariable(train_ids, f'cvfold={cv_index}_train-ids')
saveVariable(test_ids, f'cvfold={cv_index}_test-ids')
saveVariable(val_ids, f'cvfold={cv_index}_val-ids')
saveVariable(train_epoch_log,
f'cvfold={cv_index}_{model_name}-train-epoch-log')
saveVariable(val_epoch_log,
f'cvfold={cv_index}_{model_name}-val-epoch-log')
saveVariable(metric_dict,
f'cvfold={cv_index}_{model_name}-metric-dict')
saveVariable(model, f'cvfold={cv_index}_{model_name}-best')
model.load_state_dict(last_model_wts)
saveVariable(model, f'cvfold={cv_index}_{model_name}-last')
torchutils.plotEpochLog(train_epoch_log,
subfig_size=(10, 2.5),
title='Training performance',
fn=os.path.join(
fig_dir,
f'cvfold={cv_index}_train-plot.png'))
if val_epoch_log:
torchutils.plotEpochLog(val_epoch_log,
subfig_size=(10, 2.5),
title='Heldout performance',
fn=os.path.join(
fig_dir,
f'cvfold={cv_index}_val-plot.png'))
if eval_label_mapping is not None:
metric_dict = {
'Avg Loss': metrics.AverageLoss(),
'Accuracy': metrics.Accuracy(),
'Precision': metrics.Precision(),
'Recall': metrics.Recall(),
'F1': metrics.Fmeasure()
}
test_io_history = torchutils.predictSamples(
model.to(device=device),
test_loader,
criterion=criterion,
device=device,
metrics=metric_dict,
data_labeled=True,
update_model=False,
seq_as_batch=train_params['seq_as_batch'],
return_io_history=True,
label_mapping=eval_label_mapping)
if independent_signals:
test_io_history = joinSeqs(test_io_history)
metric_str = ' '.join(str(m) for m in metric_dict.values())
logger.info('[TST] ' + metric_str)
def main(out_dir=None,
data_dir=None,
prefix='trial=',
model_name=None,
gpu_dev_id=None,
batch_size=None,
learning_rate=None,
file_fn_format=None,
label_fn_format=None,
start_from=None,
stop_at=None,
model_params={},
cv_params={},
train_params={},
viz_params={},
num_disp_imgs=None,
viz_templates=None,
results_file=None,
sweep_param_name=None):
data_dir = os.path.expanduser(data_dir)
out_dir = os.path.expanduser(out_dir)
if not os.path.exists(out_dir):
os.makedirs(out_dir)
logger = utils.setupRootLogger(filename=os.path.join(out_dir, 'log.txt'))
if results_file is None:
results_file = os.path.join(out_dir, 'results.csv')
else:
results_file = os.path.expanduser(results_file)
fig_dir = os.path.join(out_dir, 'figures')
if not os.path.exists(fig_dir):
os.makedirs(fig_dir)
io_dir = os.path.join(fig_dir, 'model-io')
if not os.path.exists(io_dir):
os.makedirs(io_dir)
out_data_dir = os.path.join(out_dir, 'data')
if not os.path.exists(out_data_dir):
os.makedirs(out_data_dir)
def saveVariable(var, var_name, to_dir=out_data_dir):
utils.saveVariable(var, var_name, to_dir)
# Load data
trial_ids = utils.getUniqueIds(data_dir, prefix=prefix, to_array=True)
vocab = utils.loadVariable('vocab', data_dir)
saveVariable(vocab, 'vocab')
# Define cross-validation folds
data_loader = utils.CvDataset(trial_ids,
data_dir,
vocab=vocab,
prefix=prefix,
feature_fn_format=file_fn_format,
label_fn_format=label_fn_format)
cv_folds = utils.makeDataSplits(len(data_loader.trial_ids), **cv_params)
device = torchutils.selectDevice(gpu_dev_id)
labels_dtype = torch.long
criterion = torch.nn.CrossEntropyLoss()
metric_names = ('Loss', 'Accuracy')
def make_dataset(fns, labels, ids, batch_mode='sample', shuffle=True):
dataset = VideoDataset(fns,
labels,
device=device,
labels_dtype=labels_dtype,
seq_ids=ids,
batch_size=batch_size,
batch_mode=batch_mode)
loader = torch.utils.data.DataLoader(dataset,
batch_size=1,
shuffle=shuffle)
return dataset, loader
for cv_index, cv_fold in enumerate(cv_folds):
if start_from is not None and cv_index < start_from:
continue
if stop_at is not None and cv_index > stop_at:
break
train_data, val_data, test_data = data_loader.getFold(cv_fold)
train_set, train_loader = make_dataset(*train_data,
batch_mode='flatten',
shuffle=True)
test_set, test_loader = make_dataset(*test_data,
batch_mode='flatten',
shuffle=False)
val_set, val_loader = make_dataset(*val_data,
batch_mode='flatten',
shuffle=True)
logger.info(
f'CV fold {cv_index + 1} / {len(cv_folds)}: {len(data_loader.trial_ids)} total '
f'({len(train_set)} train, {len(val_set)} val, {len(test_set)} test)'
)
model = ImageClassifier(len(vocab), **model_params)
optimizer_ft = torch.optim.Adam(model.parameters(),
lr=learning_rate,
betas=(0.9, 0.999),
eps=1e-08,
weight_decay=0,
amsgrad=False)
lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer_ft,
step_size=1,
gamma=1.00)
train_epoch_log = collections.defaultdict(list)
val_epoch_log = collections.defaultdict(list)
metric_dict = {name: metrics.makeMetric(name) for name in metric_names}
model, last_model_wts = torchutils.trainModel(
model,
criterion,
optimizer_ft,
lr_scheduler,
train_loader,
val_loader,
device=device,
metrics=metric_dict,
train_epoch_log=train_epoch_log,
val_epoch_log=val_epoch_log,
**train_params)
# Test model
metric_dict = {name: metrics.makeMetric(name) for name in metric_names}
test_io_history = torchutils.predictSamples(
model.to(device=device),
test_loader,
criterion=criterion,
device=device,
metrics=metric_dict,
data_labeled=True,
update_model=False,
seq_as_batch=train_params['seq_as_batch'],
return_io_history=True)
metric_str = ' '.join(str(m) for m in metric_dict.values())
logger.info('[TST] ' + metric_str)
utils.writeResults(results_file,
{name: m.value
for name, m in metric_dict.items()},
sweep_param_name, model_params)
for pred_seq, score_seq, feat_seq, label_seq, batch_id in test_io_history:
prefix = f'cvfold={cv_index}_batch={batch_id}'
saveVariable(pred_seq.cpu().numpy(), f'{prefix}_pred-label-seq')
saveVariable(score_seq.cpu().numpy(), f'{prefix}_score-seq')
saveVariable(label_seq.cpu().numpy(), f'{prefix}_true-label-seq')
saveVariable(test_set.unflatten,
f'cvfold={cv_index}_test-set-unflatten')
saveVariable(model, f'cvfold={cv_index}_{model_name}-best')
if train_epoch_log:
torchutils.plotEpochLog(train_epoch_log,
subfig_size=(10, 2.5),
title='Training performance',
fn=os.path.join(
fig_dir,
f'cvfold={cv_index}_train-plot.png'))
if val_epoch_log:
torchutils.plotEpochLog(val_epoch_log,
subfig_size=(10, 2.5),
title='Heldout performance',
fn=os.path.join(
fig_dir,
f'cvfold={cv_index}_val-plot.png'))
def main(out_dir=None,
data_dir=None,
background_data_dir=None,
learn_bg_model=False,
gpu_dev_id=None,
start_from=None,
stop_at=None,
num_disp_imgs=None,
depth_bg_detection_kwargs={},
rgb_bg_detection_kwargs={}):
out_dir = os.path.expanduser(out_dir)
data_dir = os.path.expanduser(data_dir)
background_data_dir = os.path.expanduser(background_data_dir)
logger = utils.setupRootLogger(filename=os.path.join(out_dir, 'log.txt'))
fig_dir = os.path.join(out_dir, 'figures')
if not os.path.exists(fig_dir):
os.makedirs(fig_dir)
out_data_dir = os.path.join(out_dir, 'data')
if not os.path.exists(out_data_dir):
os.makedirs(out_data_dir)
def loadFromDir(var_name, dir_name):
return joblib.load(os.path.join(dir_name, f"{var_name}.pkl"))
def saveToWorkingDir(var, var_name):
joblib.dump(var, os.path.join(out_data_dir, f"{var_name}.pkl"))
trial_ids = utils.getUniqueIds(data_dir, prefix='trial=', to_array=True)
device = torchutils.selectDevice(gpu_dev_id)
camera_pose = render.camera_pose
camera_params = render.intrinsic_matrix
for seq_idx, trial_id in enumerate(trial_ids):
if start_from is not None and seq_idx < start_from:
continue
if stop_at is not None and seq_idx > stop_at:
break
trial_str = f"trial={trial_id}"
logger.info(
f"Processing video {seq_idx + 1} / {len(trial_ids)} (trial {trial_id})"
)
logger.info(" Loading data...")
try:
rgb_frame_seq = loadFromDir(f"{trial_str}_rgb-frame-seq", data_dir)
depth_frame_seq = loadFromDir(f"{trial_str}_depth-frame-seq",
data_dir)
rgb_train = loadFromDir(
f"{trial_str}_rgb-frame-seq-before-first-touch",
background_data_dir)
depth_train = loadFromDir(
f"{trial_str}_depth-frame-seq-before-first-touch",
background_data_dir)
if isinstance(depth_train, tuple) and isinstance(
depth_frame_seq, tuple):
logger.info(" Skipping video: depth frames missing")
continue
rgb_frame_seq = np.stack(tuple(
skimage.img_as_float(f) for f in rgb_frame_seq),
axis=0)
rgb_train = np.stack(tuple(
skimage.img_as_float(f) for f in rgb_train),
axis=0)
except FileNotFoundError as e:
logger.info(e)
continue
logger.info(" Removing background...")
try:
bg_mask_depth_train = loadFromDir(
f'{trial_str}_bg-mask-depth-train', out_data_dir)
bg_mask_seq_depth = loadFromDir(f'{trial_str}_bg-mask-seq-depth',
out_data_dir)
except FileNotFoundError:
bg_model_depth, bg_mask_depth_train = fitBackgroundDepth(
depth_train,
camera_params=camera_params,
camera_pose=camera_pose,
**depth_bg_detection_kwargs)
bg_mask_seq_depth = detectBackgroundDepth(
bg_model_depth,
depth_frame_seq,
camera_params=camera_params,
camera_pose=camera_pose,
**depth_bg_detection_kwargs)
__, bg_model_depth_image, __ = render.renderPlane(
bg_model_depth,
camera_pose=camera_pose,
camera_params=camera_params)
imageprocessing.displayImage(
bg_model_depth_image,
file_path=os.path.join(fig_dir,
f'{trial_str}_bg-image-depth.png'))
bg_model_rgb = fitBackgroundRgb(
np.vstack((rgb_train, rgb_frame_seq)),
np.vstack((bg_mask_depth_train, bg_mask_seq_depth)))
if learn_bg_model:
model = RgbBackgroundModel(bg_model_rgb,
update_bg=True,
device=device)
losses, metrics = model.fit(np.vstack((rgb_train, rgb_frame_seq)),
np.vstack((bg_mask_depth_train,
bg_mask_seq_depth)),
num_epochs=100)
outputs = model.forward(
torch.tensor(rgb_frame_seq, dtype=torch.float,
device=device).permute(0, -1, 1, 2))
bg_mask_seq_rgb = model.predict(outputs).cpu().numpy().squeeze()
plot_dict = {'Loss': losses, 'Accuracy': metrics},
else:
def f1(preds, targets):
true_positives = np.sum((targets == 1) * (preds == 1))
false_positives = np.sum((targets == 0) * (preds == 1))
false_negatives = np.sum((targets == 1) * (preds == 0))
precision = true_positives / (true_positives + false_positives)
recall = true_positives / (true_positives + false_negatives)
f1 = 2 * (precision * recall) / (precision + recall)
return f1
def acc(preds, targets):
matches = preds == targets
return matches.mean()
bg_dists = np.linalg.norm(rgb_frame_seq - bg_model_rgb[None, ...],
axis=-1)
thresh_vals = np.linspace(0, 1, num=50)
scores = np.array(
[acc(bg_dists < t, bg_mask_seq_depth) for t in thresh_vals])
best_index = scores.argmax()
best_thresh = thresh_vals[best_index]
bg_mask_seq_rgb = bg_dists < best_thresh
plot_dict = {'Accuracy': scores}
torchutils.plotEpochLog(plot_dict,
subfig_size=(10, 2.5),
title='Training performance',
fn=os.path.join(fig_dir,
f'{trial_str}_train-plot.png'))
logger.info(" Saving output...")
saveToWorkingDir(bg_mask_depth_train.astype(bool),
f'{trial_str}_bg-mask-depth-train')
saveToWorkingDir(bg_mask_seq_depth.astype(bool),
f'{trial_str}_bg-mask-seq-depth')
saveToWorkingDir(bg_mask_seq_rgb.astype(bool),
f'{trial_str}_bg-mask-seq-rgb')
if num_disp_imgs is not None:
if rgb_frame_seq.shape[0] > num_disp_imgs:
idxs = np.arange(rgb_frame_seq.shape[0])
np.random.shuffle(idxs)
idxs = idxs[:num_disp_imgs]
else:
idxs = slice(None, None, None)
imageprocessing.displayImages(*(rgb_frame_seq[idxs]),
*(bg_mask_seq_rgb[idxs]),
*(depth_frame_seq[idxs]),
*(bg_mask_seq_depth[idxs]),
num_rows=4,
file_path=os.path.join(
fig_dir,
f'{trial_str}_best-frames.png'))
imageprocessing.displayImage(bg_model_rgb,
file_path=os.path.join(
fig_dir,
f'{trial_str}_bg-image-rgb.png'))
def main(out_dir=None,
scores_dir=None,
preprocessed_data_dir=None,
keyframe_model_name=None,
subsample_period=None,
window_size=None,
corpus_name=None,
default_annotator=None,
cv_scheme=None,
max_trials_per_fold=None,
model_name=None,
numeric_backend=None,
gpu_dev_id=None,
visualize=False,
model_config={},
camera_params_config={}):
out_dir = os.path.expanduser(out_dir)
scores_dir = os.path.expanduser(scores_dir)
preprocessed_data_dir = os.path.expanduser(preprocessed_data_dir)
m.set_backend('numpy')
def loadFromWorkingDir(var_name):
return joblib.load(os.path.join(scores_dir, f"{var_name}.pkl"))
def saveToWorkingDir(var, var_name):
joblib.dump(var, os.path.join(out_dir, f"{var_name}.pkl"))
# Load camera parameters from external file and add them to model config kwargs
model_config['init_kwargs'].update(
render.loadCameraParams(**camera_params_config, as_dict=True))
trial_ids = joblib.load(
os.path.join(preprocessed_data_dir, 'trial_ids.pkl'))
corpus = duplocorpus.DuploCorpus(corpus_name)
assembly_seqs = tuple(
labels.parseLabelSeq(
corpus.readLabels(trial_id, default_annotator)[0])
for trial_id in trial_ids)
logger.info(f"Selecting keyframes...")
keyframe_idx_seqs = []
rgb_keyframe_seqs = []
depth_keyframe_seqs = []
seg_keyframe_seqs = []
background_keyframe_seqs = []
assembly_keyframe_seqs = []
for seq_idx, trial_id in enumerate(trial_ids):
trial_str = f"trial-{trial_id}"
rgb_frame_seq = loadFromWorkingDir(f'{trial_str}_rgb-frame-seq')
depth_frame_seq = loadFromWorkingDir(f'{trial_str}_depth-frame-seq')
segment_seq = loadFromWorkingDir(f'{trial_str}_segment-seq')
frame_scores = loadFromWorkingDir(f'{trial_str}_frame-scores')
background_plane_seq = loadFromWorkingDir(
f'{trial_str}_background-plane-seq')
assembly_seq = assembly_seqs[seq_idx]
# FIXME: Get the real frame index numbers instead of approximating
assembly_seq[-1].end_idx = len(rgb_frame_seq) * subsample_period
keyframe_idxs = videoprocessing.selectSegmentKeyframes(
frame_scores, score_thresh=0, prepend_first=True)
selectKeyframes = functools.partial(utils.select, keyframe_idxs)
rgb_keyframe_seq = selectKeyframes(rgb_frame_seq)
depth_keyframe_seq = selectKeyframes(depth_frame_seq)
seg_keyframe_seq = selectKeyframes(segment_seq)
background_keyframe_seq = selectKeyframes(background_plane_seq)
# FIXME: Get the real frame index numbers instead of approximating
keyframe_idxs_orig = keyframe_idxs * subsample_period
assembly_keyframe_seq = labels.resampleStateSeq(
keyframe_idxs_orig, assembly_seq)
# Store all keyframe sequences in memory
keyframe_idx_seqs.append(keyframe_idxs)
rgb_keyframe_seqs.append(rgb_keyframe_seq)
depth_keyframe_seqs.append(depth_keyframe_seq)
seg_keyframe_seqs.append(seg_keyframe_seq)
background_keyframe_seqs.append(background_keyframe_seq)
assembly_keyframe_seqs.append(assembly_keyframe_seq)
# Split into train and test sets
if cv_scheme == 'leave one out':
num_seqs = len(trial_ids)
cv_folds = []
for i in range(num_seqs):
test_fold = (i, )
train_fold = tuple(range(0, i)) + tuple(range(i + 1, num_seqs))
cv_folds.append((train_fold, test_fold))
elif cv_scheme == 'train on child':
child_corpus = duplocorpus.DuploCorpus('child')
child_trial_ids = utils.loadVariable('trial_ids',
'preprocess-all-data', 'child')
child_assembly_seqs = [
labels.parseLabelSeq(
child_corpus.readLabels(trial_id, 'Cathryn')[0])
for trial_id in child_trial_ids
]
num_easy = len(assembly_keyframe_seqs)
num_child = len(child_assembly_seqs)
cv_folds = [(tuple(range(num_easy, num_easy + num_child)),
tuple(range(num_easy)))]
assembly_keyframe_seqs = assembly_keyframe_seqs + child_assembly_seqs
rgb_keyframe_seqs = tuple(
tuple(
imageprocessing.saturateImage(rgb_image,
background_mask=segment_image == 0)
for rgb_image, segment_image in zip(rgb_frame_seq, seg_frame_seq))
for rgb_frame_seq, seg_frame_seq in zip(rgb_keyframe_seqs,
seg_keyframe_seqs))
depth_keyframe_seqs = tuple(
tuple(depth_image.astype(float) for depth_image in depth_frame_seq)
for depth_frame_seq in depth_keyframe_seqs)
device = torchutils.selectDevice(gpu_dev_id)
m.set_backend('torch')
m.set_default_device(device)
assembly_keyframe_seqs = tuple(
tuple(a.to(device=device, in_place=False) for a in seq)
for seq in assembly_keyframe_seqs)
assembly_seqs = tuple(
tuple(a.to(device=device, in_place=False) for a in seq)
for seq in assembly_seqs)
rgb_keyframe_seqs = tuple(
tuple(m.np.array(frame, dtype=torch.float) for frame in rgb_frame_seq)
for rgb_frame_seq in rgb_keyframe_seqs)
depth_keyframe_seqs = tuple(
tuple(
m.np.array(frame, dtype=torch.float) for frame in depth_frame_seq)
for depth_frame_seq in depth_keyframe_seqs)
seg_keyframe_seqs = tuple(
tuple(m.np.array(frame, dtype=torch.int) for frame in seg_frame_seq)
for seg_frame_seq in seg_keyframe_seqs)
num_cv_folds = len(cv_folds)
saveToWorkingDir(cv_folds, f'cv-folds')
for fold_index, (train_idxs, test_idxs) in enumerate(cv_folds):
logger.info(f"CV FOLD {fold_index + 1} / {num_cv_folds}")
# Initialize and train model
utils.validateCvFold(train_idxs, test_idxs)
selectTrain = functools.partial(utils.select, train_idxs)
train_assembly_seqs = selectTrain(assembly_keyframe_seqs)
model = getattr(models, model_name)(**model_config['init_kwargs'])
logger.info(
f" Training {model_name} on {len(train_idxs)} sequences...")
model.fit(train_assembly_seqs, **model_config['fit_kwargs'])
logger.info(
f' Model trained on {model.num_states} unique assembly states')
# saveToWorkingDir(model, f'model-fold{fold_index}')
# Decode on the test set
selectTest = functools.partial(utils.select, test_idxs)
test_trial_ids = selectTest(trial_ids)
test_rgb_keyframe_seqs = selectTest(rgb_keyframe_seqs)
test_depth_keyframe_seqs = selectTest(depth_keyframe_seqs)
test_seg_keyframe_seqs = selectTest(seg_keyframe_seqs)
test_background_keyframe_seqs = selectTest(background_keyframe_seqs)
test_assembly_keyframe_seqs = selectTest(assembly_keyframe_seqs)
test_assembly_seqs = selectTest(assembly_seqs)
logger.info(f" Testing model on {len(test_idxs)} sequences...")
for i, trial_id in enumerate(test_trial_ids):
if max_trials_per_fold is not None and i >= max_trials_per_fold:
break
rgb_frame_seq = test_rgb_keyframe_seqs[i]
depth_frame_seq = test_depth_keyframe_seqs[i]
seg_frame_seq = test_seg_keyframe_seqs[i]
background_plane_seq = test_background_keyframe_seqs[i]
true_assembly_seq = test_assembly_keyframe_seqs[i]
true_assembly_seq_orig = test_assembly_seqs[i]
rgb_background_seq, depth_background_seq = utils.batchProcess(
model.renderPlane, background_plane_seq, unzip=True)
logger.info(f' Decoding video {trial_id}...')
start_time = time.process_time()
out = model.predictSeq(rgb_frame_seq, depth_frame_seq,
seg_frame_seq, rgb_background_seq,
depth_background_seq,
**model_config['decode_kwargs'])
pred_assembly_seq, pred_idx_seq, max_log_probs, log_likelihoods, poses_seq = out
end_time = time.process_time()
logger.info(utils.makeProcessTimeStr(end_time - start_time))
num_correct, num_total = metrics.numberCorrect(
true_assembly_seq, pred_assembly_seq)
logger.info(f' ACCURACY: {num_correct} / {num_total}')
num_correct, num_total = metrics.numberCorrect(
true_assembly_seq, pred_assembly_seq, ignore_empty_true=True)
logger.info(f' RECALL: {num_correct} / {num_total}')
num_correct, num_total = metrics.numberCorrect(
true_assembly_seq, pred_assembly_seq, ignore_empty_pred=True)
logger.info(f' PRECISION: {num_correct} / {num_total}')
# Save intermediate results
logger.info(f"Saving output...")
saveToWorkingDir(segment_seq, f'segment_seq-{trial_id}')
saveToWorkingDir(true_assembly_seq_orig,
f'true_state_seq_orig-{trial_id}')
saveToWorkingDir(true_assembly_seq, f'true_state_seq-{trial_id}')
saveToWorkingDir(pred_assembly_seq, f'pred_state_seq-{trial_id}')
saveToWorkingDir(poses_seq, f'poses_seq-{trial_id}')
saveToWorkingDir(background_plane_seq,
f'background_plane_seq-{trial_id}')
saveToWorkingDir(max_log_probs, f'max_log_probs-{trial_id}')
saveToWorkingDir(log_likelihoods, f'log_likelihoods-{trial_id}')
# Save figures
if visualize:
rgb_rendered_seq, depth_rendered_seq, label_rendered_seq = utils.batchProcess(
model.renderScene,
pred_assembly_seq,
poses_seq,
rgb_background_seq,
depth_background_seq,
unzip=True,
static_kwargs={'as_numpy': True})
if utils.in_ipython_console():
file_path = None
else:
trial_str = f"trial-{trial_id}"
file_path = os.path.join(out_dir,
f'{trial_str}_best-frames.png')
rgb_frame_seq = tuple(img.cpu().numpy()
for img in rgb_frame_seq)
imageprocessing.displayImages(*rgb_frame_seq,
*rgb_rendered_seq,
num_rows=2,
file_path=file_path)
def main(
out_dir=None, modalities=['rgb', 'imu'], gpu_dev_id=None, plot_io=None,
rgb_data_dir=None, rgb_attributes_dir=None, imu_data_dir=None, imu_attributes_dir=None):
out_dir = os.path.expanduser(out_dir)
rgb_data_dir = os.path.expanduser(rgb_data_dir)
rgb_attributes_dir = os.path.expanduser(rgb_attributes_dir)
imu_data_dir = os.path.expanduser(imu_data_dir)
imu_attributes_dir = os.path.expanduser(imu_attributes_dir)
logger = utils.setupRootLogger(filename=os.path.join(out_dir, 'log.txt'))
fig_dir = os.path.join(out_dir, 'figures')
if not os.path.exists(fig_dir):
os.makedirs(fig_dir)
out_data_dir = os.path.join(out_dir, 'data')
if not os.path.exists(out_data_dir):
os.makedirs(out_data_dir)
# Load data
if modalities == ['rgb']:
trial_ids = utils.getUniqueIds(rgb_data_dir, prefix='trial=', to_array=True)
logger.info(f"Processing {len(trial_ids)} videos")
else:
rgb_trial_ids = utils.getUniqueIds(rgb_data_dir, prefix='trial=', to_array=True)
imu_trial_ids = utils.getUniqueIds(imu_data_dir, prefix='trial=', to_array=True)
trial_ids = np.array(sorted(set(rgb_trial_ids.tolist()) & set(imu_trial_ids.tolist())))
logger.info(
f"Processing {len(trial_ids)} videos common to "
f"RGB ({len(rgb_trial_ids)} total) and IMU ({len(imu_trial_ids)} total)"
)
device = torchutils.selectDevice(gpu_dev_id)
dataset = FusionDataset(
trial_ids, rgb_attributes_dir, rgb_data_dir, imu_attributes_dir, imu_data_dir,
device=device, modalities=modalities,
)
utils.saveMetadata(dataset.metadata, out_data_dir)
utils.saveVariable(dataset.vocab, 'vocab', out_data_dir)
for i, trial_id in enumerate(trial_ids):
logger.info(f"Processing sequence {trial_id}...")
true_label_seq = dataset.loadTargets(trial_id)
attribute_feats = dataset.loadInputs(trial_id)
# (Process the samples here if we need to)
attribute_feats = attribute_feats.cpu().numpy()
true_label_seq = true_label_seq.cpu().numpy()
trial_prefix = f"trial={trial_id}"
utils.saveVariable(attribute_feats, f'{trial_prefix}_feature-seq', out_data_dir)
utils.saveVariable(true_label_seq, f'{trial_prefix}_label-seq', out_data_dir)
if plot_io:
fn = os.path.join(fig_dir, f'{trial_prefix}.png')
utils.plot_array(
attribute_feats.T,
(true_label_seq,),
('gt',),
fn=fn
)
def main(out_dir=None,
data_dir=None,
model_name=None,
part_symmetries=None,
gpu_dev_id=None,
batch_size=None,
learning_rate=None,
model_params={},
cv_params={},
train_params={},
viz_params={},
plot_predictions=None,
results_file=None,
sweep_param_name=None):
if part_symmetries is None:
part_symmetries = {
'beam_side': ('backbeam_hole_1', 'backbeam_hole_2',
'frontbeam_hole_1', 'frontbeam_hole_2'),
'beam_top': ('backbeam_hole_3', 'frontbeam_hole_3'),
'backrest': ('backrest_hole_1', 'backrest_hole_2')
}
data_dir = os.path.expanduser(data_dir)
out_dir = os.path.expanduser(out_dir)
if not os.path.exists(out_dir):
os.makedirs(out_dir)
logger = utils.setupRootLogger(filename=os.path.join(out_dir, 'log.txt'))
if results_file is None:
results_file = os.path.join(out_dir, 'results.csv')
else:
results_file = os.path.expanduser(results_file)
fig_dir = os.path.join(out_dir, 'figures')
if not os.path.exists(fig_dir):
os.makedirs(fig_dir)
out_data_dir = os.path.join(out_dir, 'data')
if not os.path.exists(out_data_dir):
os.makedirs(out_data_dir)
def saveVariable(var, var_name):
joblib.dump(var, os.path.join(out_data_dir, f'{var_name}.pkl'))
def loadAll(seq_ids, var_name, data_dir):
def loadOne(seq_id):
fn = os.path.join(data_dir, f'trial={seq_id}_{var_name}')
return joblib.load(fn)
return tuple(map(loadOne, seq_ids))
# Load vocab
with open(os.path.join(data_dir, "part-vocab.yaml"), 'rt') as f:
link_vocab = yaml.safe_load(f)
assembly_vocab = joblib.load(os.path.join(data_dir, 'assembly-vocab.pkl'))
# Load data
trial_ids = utils.getUniqueIds(data_dir, prefix='trial=')
feature_seqs = loadAll(trial_ids, 'feature-seq.pkl', data_dir)
label_seqs = loadAll(trial_ids, 'label-seq.pkl', data_dir)
if part_symmetries:
# Construct equivalence classes from vocab
eq_classes, assembly_eq_classes, eq_class_vocab = makeEqClasses(
assembly_vocab, part_symmetries)
lib_assembly.writeAssemblies(
os.path.join(fig_dir, 'eq-class-vocab.txt'), eq_class_vocab)
label_seqs = tuple(assembly_eq_classes[label_seq]
for label_seq in label_seqs)
saveVariable(eq_class_vocab, 'assembly-vocab')
else:
eq_classes = None
def impute_nan(input_seq):
input_is_nan = np.isnan(input_seq)
logger.info(f"{input_is_nan.sum()} NaN elements")
input_seq[input_is_nan] = 0 # np.nanmean(input_seq)
return input_seq
# feature_seqs = tuple(map(impute_nan, feature_seqs))
for trial_id, label_seq, feat_seq in zip(trial_ids, label_seqs,
feature_seqs):
saveVariable(feat_seq, f"trial={trial_id}_feature-seq")
saveVariable(label_seq, f"trial={trial_id}_label-seq")
device = torchutils.selectDevice(gpu_dev_id)
# Define cross-validation folds
dataset_size = len(trial_ids)
cv_folds = utils.makeDataSplits(dataset_size, **cv_params)
def getSplit(split_idxs):
split_data = tuple(
tuple(s[i] for i in split_idxs)
for s in (feature_seqs, label_seqs, trial_ids))
return split_data
for cv_index, cv_splits in enumerate(cv_folds):
train_data, val_data, test_data = tuple(map(getSplit, cv_splits))
train_feats, train_labels, train_ids = train_data
train_set = torchutils.SequenceDataset(train_feats,
train_labels,
device=device,
labels_dtype=torch.long,
seq_ids=train_ids,
transpose_data=True)
train_loader = torch.utils.data.DataLoader(train_set,
batch_size=batch_size,
shuffle=True)
test_feats, test_labels, test_ids = test_data
test_set = torchutils.SequenceDataset(test_feats,
test_labels,
device=device,
labels_dtype=torch.long,
seq_ids=test_ids,
transpose_data=True)
test_loader = torch.utils.data.DataLoader(test_set,
batch_size=batch_size,
shuffle=False)
val_feats, val_labels, val_ids = val_data
val_set = torchutils.SequenceDataset(val_feats,
val_labels,
device=device,
labels_dtype=torch.long,
seq_ids=val_ids,
transpose_data=True)
val_loader = torch.utils.data.DataLoader(val_set,
batch_size=batch_size,
shuffle=True)
logger.info(
f'CV fold {cv_index + 1} / {len(cv_folds)}: {len(trial_ids)} total '
f'({len(train_ids)} train, {len(val_ids)} val, {len(test_ids)} test)'
)
input_dim = train_set.num_obsv_dims
output_dim = train_set.num_label_types
if model_name == 'linear':
model = torchutils.LinearClassifier(
input_dim, output_dim, **model_params).to(device=device)
elif model_name == 'dummy':
model = DummyClassifier(input_dim, output_dim, **model_params)
elif model_name == 'AssemblyClassifier':
model = AssemblyClassifier(assembly_vocab,
link_vocab,
eq_classes=eq_classes,
**model_params)
else:
raise AssertionError()
criterion = torch.nn.CrossEntropyLoss()
if model_name != 'dummy':
train_epoch_log = collections.defaultdict(list)
val_epoch_log = collections.defaultdict(list)
metric_dict = {
'Avg Loss': metrics.AverageLoss(),
'Accuracy': metrics.Accuracy(),
'Precision': metrics.Precision(),
'Recall': metrics.Recall(),
'F1': metrics.Fmeasure()
}
optimizer_ft = torch.optim.Adam(model.parameters(),
lr=learning_rate,
betas=(0.9, 0.999),
eps=1e-08,
weight_decay=0,
amsgrad=False)
lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer_ft,
step_size=1,
gamma=1.00)
model, last_model_wts = torchutils.trainModel(
model,
criterion,
optimizer_ft,
lr_scheduler,
train_loader,
val_loader,
device=device,
metrics=metric_dict,
train_epoch_log=train_epoch_log,
val_epoch_log=val_epoch_log,
**train_params)
logger.info(f'scale={float(model._scale)}')
logger.info(f'alpha={float(model._alpha)}')
# Test model
metric_dict = {
'Avg Loss': metrics.AverageLoss(),
'Accuracy': metrics.Accuracy(),
'Precision': metrics.Precision(),
'Recall': metrics.Recall(),
'F1': metrics.Fmeasure()
}
test_io_history = torchutils.predictSamples(
model.to(device=device),
test_loader,
criterion=criterion,
device=device,
metrics=metric_dict,
data_labeled=True,
update_model=False,
seq_as_batch=train_params['seq_as_batch'],
return_io_history=True)
metric_str = ' '.join(str(m) for m in metric_dict.values())
logger.info('[TST] ' + metric_str)
d = {k: v.value for k, v in metric_dict.items()}
utils.writeResults(results_file, d, sweep_param_name, model_params)
if plot_predictions:
io_fig_dir = os.path.join(fig_dir, 'model-io')
if not os.path.exists(io_fig_dir):
os.makedirs(io_fig_dir)
label_names = ('gt', 'pred')
preds, scores, inputs, gt_labels, ids = zip(*test_io_history)
for batch in test_io_history:
batch = tuple(
x.cpu().numpy() if isinstance(x, torch.Tensor) else x
for x in batch)
for preds, _, inputs, gt_labels, seq_id in zip(*batch):
fn = os.path.join(io_fig_dir,
f"trial={seq_id}_model-io.png")
utils.plot_array(inputs.sum(axis=-1), (gt_labels, preds),
label_names,
fn=fn,
**viz_params)
def saveTrialData(pred_seq, score_seq, feat_seq, label_seq, trial_id):
saveVariable(pred_seq, f'trial={trial_id}_pred-label-seq')
saveVariable(score_seq, f'trial={trial_id}_score-seq')
saveVariable(label_seq, f'trial={trial_id}_true-label-seq')
for batch in test_io_history:
batch = tuple(x.cpu().numpy() if isinstance(x, torch.Tensor) else x
for x in batch)
for io in zip(*batch):
saveTrialData(*io)
saveVariable(train_ids, f'cvfold={cv_index}_train-ids')
saveVariable(test_ids, f'cvfold={cv_index}_test-ids')
saveVariable(val_ids, f'cvfold={cv_index}_val-ids')
saveVariable(train_epoch_log,
f'cvfold={cv_index}_{model_name}-train-epoch-log')
saveVariable(val_epoch_log,
f'cvfold={cv_index}_{model_name}-val-epoch-log')
saveVariable(metric_dict,
f'cvfold={cv_index}_{model_name}-metric-dict')
saveVariable(model, f'cvfold={cv_index}_{model_name}-best')
train_fig_dir = os.path.join(fig_dir, 'train-plots')
if not os.path.exists(train_fig_dir):
os.makedirs(train_fig_dir)
if train_epoch_log:
torchutils.plotEpochLog(train_epoch_log,
subfig_size=(10, 2.5),
title='Training performance',
fn=os.path.join(
train_fig_dir,
f'cvfold={cv_index}_train-plot.png'))
if val_epoch_log:
torchutils.plotEpochLog(val_epoch_log,
subfig_size=(10, 2.5),
title='Heldout performance',
fn=os.path.join(
train_fig_dir,
f'cvfold={cv_index}_val-plot.png'))
def main(out_dir=None,
data_dir=None,
segs_dir=None,
pretrained_model_dir=None,
model_name=None,
gpu_dev_id=None,
batch_size=None,
learning_rate=None,
start_from=None,
stop_at=None,
model_params={},
cv_params={},
train_params={},
viz_params={},
num_disp_imgs=None,
viz_templates=None,
results_file=None,
sweep_param_name=None):
data_dir = os.path.expanduser(data_dir)
segs_dir = os.path.expanduser(segs_dir)
pretrained_model_dir = os.path.expanduser(pretrained_model_dir)
out_dir = os.path.expanduser(out_dir)
if not os.path.exists(out_dir):
os.makedirs(out_dir)
logger = utils.setupRootLogger(filename=os.path.join(out_dir, 'log.txt'))
if results_file is None:
results_file = os.path.join(out_dir, 'results.csv')
else:
results_file = os.path.expanduser(results_file)
fig_dir = os.path.join(out_dir, 'figures')
if not os.path.exists(fig_dir):
os.makedirs(fig_dir)
io_dir = os.path.join(fig_dir, 'model-io')
if not os.path.exists(io_dir):
os.makedirs(io_dir)
out_data_dir = os.path.join(out_dir, 'data')
if not os.path.exists(out_data_dir):
os.makedirs(out_data_dir)
def saveVariable(var, var_name, to_dir=out_data_dir):
utils.saveVariable(var, var_name, to_dir)
# Load data
trial_ids = utils.getUniqueIds(data_dir, prefix='trial=', to_array=True)
vocab = utils.loadVariable('vocab', pretrained_model_dir)
parts_vocab = utils.loadVariable('parts-vocab', pretrained_model_dir)
edge_labels = utils.loadVariable('part-labels', pretrained_model_dir)
saveVariable(vocab, 'vocab')
saveVariable(parts_vocab, 'parts-vocab')
saveVariable(edge_labels, 'part-labels')
# Define cross-validation folds
data_loader = VideoLoader(trial_ids,
data_dir,
segs_dir,
vocab=vocab,
label_fn_format='assembly-seq')
cv_folds = utils.makeDataSplits(len(data_loader.trial_ids), **cv_params)
Dataset = sim2real.BlocksConnectionDataset
device = torchutils.selectDevice(gpu_dev_id)
label_dtype = torch.long
labels_dtype = torch.long # FIXME
criterion = torch.nn.CrossEntropyLoss()
def make_dataset(labels, ids, batch_mode='sample', shuffle=True):
dataset = Dataset(vocab,
edge_labels,
label_dtype,
data_loader.loadData,
labels,
device=device,
labels_dtype=labels_dtype,
seq_ids=ids,
batch_size=batch_size,
batch_mode=batch_mode)
loader = torch.utils.data.DataLoader(dataset,
batch_size=1,
shuffle=shuffle)
return dataset, loader
for cv_index, cv_fold in enumerate(cv_folds):
if start_from is not None and cv_index < start_from:
continue
if stop_at is not None and cv_index > stop_at:
break
train_data, val_data, test_data = data_loader.getFold(cv_fold)
train_set, train_loader = make_dataset(*train_data,
batch_mode='sample',
shuffle=True)
test_set, test_loader = make_dataset(*test_data,
batch_mode='flatten',
shuffle=False)
val_set, val_loader = make_dataset(*val_data,
batch_mode='sample',
shuffle=True)
logger.info(
f'CV fold {cv_index + 1} / {len(cv_folds)}: {len(data_loader.trial_ids)} total '
f'({len(train_set)} train, {len(val_set)} val, {len(test_set)} test)'
)
logger.info(
f"Class freqs (train): {np.squeeze(train_set.class_freqs)}")
logger.info(f"Class freqs (val): {np.squeeze(val_set.class_freqs)}")
logger.info(f"Class freqs (test): {np.squeeze(test_set.class_freqs)}")
if model_name == 'template':
model = sim2real.AssemblyClassifier(vocab, **model_params)
elif model_name == 'pretrained':
pretrained_model = utils.loadVariable("cvfold=0_model-best",
pretrained_model_dir)
model = sim2real.SceneClassifier(pretrained_model, **model_params)
metric_names = ('Loss', 'Accuracy', 'Precision', 'Recall', 'F1')
criterion = torch.nn.CrossEntropyLoss()
# criterion = torchutils.BootstrappedCriterion(
# 0.25, base_criterion=torch.nn.functional.cross_entropy,
# )
else:
raise AssertionError()
optimizer_ft = torch.optim.Adam(model.parameters(),
lr=learning_rate,
betas=(0.9, 0.999),
eps=1e-08,
weight_decay=0,
amsgrad=False)
lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer_ft,
step_size=1,
gamma=1.00)
train_epoch_log = collections.defaultdict(list)
val_epoch_log = collections.defaultdict(list)
metric_dict = {name: metrics.makeMetric(name) for name in metric_names}
model, last_model_wts = torchutils.trainModel(
model,
criterion,
optimizer_ft,
lr_scheduler,
train_loader,
val_loader,
device=device,
metrics=metric_dict,
train_epoch_log=train_epoch_log,
val_epoch_log=val_epoch_log,
**train_params)
# Test model
metric_dict = {name: metrics.makeMetric(name) for name in metric_names}
test_io_history = torchutils.predictSamples(
model.to(device=device),
test_loader,
criterion=criterion,
device=device,
metrics=metric_dict,
data_labeled=True,
update_model=False,
seq_as_batch=train_params['seq_as_batch'],
return_io_history=True)
metric_str = ' '.join(str(m) for m in metric_dict.values())
logger.info('[TST] ' + metric_str)
utils.writeResults(results_file,
{name: m.value
for name, m in metric_dict.items()},
sweep_param_name, model_params)
for pred_seq, score_seq, feat_seq, label_seq, batch_id in test_io_history:
prefix = f'cvfold={cv_index}_batch={batch_id}'
saveVariable(pred_seq.cpu().numpy(), f'{prefix}_pred-label-seq')
saveVariable(score_seq.cpu().numpy(), f'{prefix}_score-seq')
saveVariable(label_seq.cpu().numpy(), f'{prefix}_true-label-seq')
saveVariable(test_set.unflatten,
f'cvfold={cv_index}_test-set-unflatten')
saveVariable(model, f'cvfold={cv_index}_{model_name}-best')
if train_epoch_log:
torchutils.plotEpochLog(train_epoch_log,
subfig_size=(10, 2.5),
title='Training performance',
fn=os.path.join(
fig_dir,
f'cvfold={cv_index}_train-plot.png'))
if val_epoch_log:
torchutils.plotEpochLog(val_epoch_log,
subfig_size=(10, 2.5),
title='Heldout performance',
fn=os.path.join(
fig_dir,
f'cvfold={cv_index}_val-plot.png'))
if model_name == 'pretrained' and num_disp_imgs is not None:
cvfold_dir = os.path.join(io_dir, f'cvfold={cv_index}')
if not os.path.exists(cvfold_dir):
os.makedirs(cvfold_dir)
model.plotBatches(test_io_history,
cvfold_dir,
images_per_fig=num_disp_imgs,
dataset=test_set)
if model_name == 'template' and num_disp_imgs is not None:
io_dir = os.path.join(fig_dir, 'model-io')
if not os.path.exists(io_dir):
os.makedirs(io_dir)
plot_topk(model, test_io_history, num_disp_imgs,
os.path.join(io_dir, f"cvfold={cv_index}.png"))
if viz_templates:
sim2real.viz_model_params(model, templates_dir=None)
def main(out_dir=None,
data_dir=None,
gpu_dev_id=None,
batch_size=None,
start_from=None,
stop_at=None,
num_disp_imgs=None):
out_dir = os.path.expanduser(out_dir)
data_dir = os.path.expanduser(data_dir)
logger = utils.setupRootLogger(filename=os.path.join(out_dir, 'log.txt'))
fig_dir = os.path.join(out_dir, 'figures')
if not os.path.exists(fig_dir):
os.makedirs(fig_dir)
out_data_dir = os.path.join(out_dir, 'data')
if not os.path.exists(out_data_dir):
os.makedirs(out_data_dir)
def loadFromDir(var_name, dir_name):
return joblib.load(os.path.join(dir_name, f"{var_name}.pkl"))
def saveToWorkingDir(var, var_name):
joblib.dump(var, os.path.join(out_data_dir, f"{var_name}.pkl"))
trial_ids = utils.getUniqueIds(data_dir, prefix='trial=', to_array=True)
device = torchutils.selectDevice(gpu_dev_id)
for seq_idx, trial_id in enumerate(trial_ids):
if start_from is not None and seq_idx < start_from:
continue
if stop_at is not None and seq_idx > stop_at:
break
trial_str = f"trial={trial_id}"
logger.info(
f"Processing video {seq_idx + 1} / {len(trial_ids)} (trial {trial_id})"
)
logger.info(" Loading data...")
try:
rgb_frame_seq = loadFromDir(f"{trial_str}_rgb-frame-seq", data_dir)
rgb_frame_seq = np.stack(tuple(
skimage.img_as_float(f) for f in rgb_frame_seq),
axis=0)
except FileNotFoundError as e:
logger.info(e)
continue
logger.info(" Detecting objects...")
model = torchvision.models.detection.maskrcnn_resnet50_fpn(
pretrained=True)
model = model.to(device=device)
model.device = device
model.eval()
inputs = np.moveaxis(rgb_frame_seq, 3, 1)
if batch_size is None:
batch_size = inputs.shape[0]
def detectBatch(batch_index):
start = batch_size * batch_index
end = start + batch_size
in_batch = torch.tensor(inputs[start:end], dtype=torch.float)
out_batches = detectCategories(model, in_batch)
return tuple(batch.numpy().squeeze(axis=1)
for batch in out_batches)
num_batches = math.ceil(inputs.shape[0] / batch_size)
person_mask_seq, bg_mask_seq = map(
np.vstack, zip(*(detectBatch(i) for i in range(num_batches))))
person_mask_seq = person_mask_seq.astype(bool)
logger.info(" Saving output...")
saveToWorkingDir(person_mask_seq, f'{trial_str}_person-mask-seq')
if num_disp_imgs is not None:
if rgb_frame_seq.shape[0] > num_disp_imgs:
idxs = np.arange(rgb_frame_seq.shape[0])
np.random.shuffle(idxs)
idxs = idxs[:num_disp_imgs]
else:
idxs = slice(None, None, None)
imageprocessing.displayImages(*(rgb_frame_seq[idxs]),
*(person_mask_seq[idxs]),
*(bg_mask_seq[idxs]),
num_rows=3,
file_path=os.path.join(
fig_dir,
f'{trial_str}_best-frames.png'))