Spaces:
Sleeping
Sleeping
Fix OOM crash when training KNN with large k by chunking full-image prediction; make the labeling-accuracy button actually jump to the Resultats tab
Browse files
app.py
CHANGED
|
@@ -56,6 +56,14 @@ CLASS_REMAP_LUT = np.array([0, 1, 4, 7, 5, 6, 2, 3], dtype=np.uint8)
|
|
| 56 |
|
| 57 |
BASE_DIR = os.path.dirname(os.path.abspath(__file__))
|
| 58 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 59 |
# โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
|
| 60 |
# Data loading (once at startup)
|
| 61 |
# โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
|
|
@@ -92,6 +100,15 @@ DATA = load_data()
|
|
| 92 |
# โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
|
| 93 |
# Model training (from the student's own labels + real spectral bands)
|
| 94 |
# โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 95 |
def train_and_predict(student_labels, k):
|
| 96 |
"""Train a KNN classifier on the student's 23 labeled polygons (using the
|
| 97 |
real spectral bands), then evaluate it only at the Ground Truth pixel
|
|
@@ -122,7 +139,7 @@ def train_and_predict(student_labels, k):
|
|
| 122 |
# with a known Ground Truth. We can't evaluate accuracy outside GT locations
|
| 123 |
# (no known answer there) but we can still show what the model predicted.
|
| 124 |
H, W, n_bands = bands.shape
|
| 125 |
-
pred_full =
|
| 126 |
|
| 127 |
return model_oa, model_matrix, pred_gt, y_gt, pred_full
|
| 128 |
|
|
@@ -153,7 +170,7 @@ def train_and_predict_logreg(student_labels):
|
|
| 153 |
model_oa = model_matrix.diagonal().sum() / model_matrix.sum()
|
| 154 |
|
| 155 |
H, W, n_bands = bands.shape
|
| 156 |
-
pred_full =
|
| 157 |
|
| 158 |
return model_oa, model_matrix, pred_gt, y_gt, pred_full
|
| 159 |
|
|
@@ -183,7 +200,7 @@ def train_and_predict_rf(student_labels):
|
|
| 183 |
model_oa = model_matrix.diagonal().sum() / model_matrix.sum()
|
| 184 |
|
| 185 |
H, W, n_bands = bands.shape
|
| 186 |
-
pred_full =
|
| 187 |
|
| 188 |
return model_oa, model_matrix, pred_gt, y_gt, pred_full
|
| 189 |
|
|
@@ -687,6 +704,7 @@ sur les pixels de vรฉritรฉ terrain (indรฉpendants des polygones d'entraรฎnement)
|
|
| 687 |
gr.update(value=table, visible=True),
|
| 688 |
gr.update(visible=True),
|
| 689 |
gr.update(value=student_fig),
|
|
|
|
| 690 |
)
|
| 691 |
|
| 692 |
def on_train(current_labels, k):
|
|
@@ -879,6 +897,7 @@ sur les pixels de vรฉritรฉ terrain (indรฉpendants des polygones d'entraรฎnement)
|
|
| 879 |
results_tbl,
|
| 880 |
row_plots_student,
|
| 881 |
student_matrix_plot,
|
|
|
|
| 882 |
],
|
| 883 |
)
|
| 884 |
|
|
|
|
| 56 |
|
| 57 |
BASE_DIR = os.path.dirname(os.path.abspath(__file__))
|
| 58 |
|
| 59 |
+
# Chunk size for predicting the full ~4.6M-pixel image. KNN's predict() builds
|
| 60 |
+
# a (n_queries, k) neighbor distance/index array โ at k=100 over the full
|
| 61 |
+
# image that's ~4.6M x 100 x 8 bytes x 2 โ 7.4GB in one call, which (combined
|
| 62 |
+
# with n_jobs=-1 worker overhead) exceeded the Space's 16GB memory limit and
|
| 63 |
+
# crashed it. Predicting in bounded chunks keeps peak memory constant
|
| 64 |
+
# regardless of k or image size.
|
| 65 |
+
FULL_PREDICT_CHUNK = 200_000
|
| 66 |
+
|
| 67 |
# โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
|
| 68 |
# Data loading (once at startup)
|
| 69 |
# โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
|
|
|
|
| 100 |
# โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
|
| 101 |
# Model training (from the student's own labels + real spectral bands)
|
| 102 |
# โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
|
| 103 |
+
def predict_full_image(clf, X_full, H, W):
|
| 104 |
+
"""Predict on the whole image in bounded chunks (see FULL_PREDICT_CHUNK)
|
| 105 |
+
instead of one call over all ~4.6M pixels at once."""
|
| 106 |
+
preds = np.empty(X_full.shape[0], dtype=np.uint8)
|
| 107 |
+
for start in range(0, X_full.shape[0], FULL_PREDICT_CHUNK):
|
| 108 |
+
end = start + FULL_PREDICT_CHUNK
|
| 109 |
+
preds[start:end] = clf.predict(X_full[start:end])
|
| 110 |
+
return preds.reshape(H, W)
|
| 111 |
+
|
| 112 |
def train_and_predict(student_labels, k):
|
| 113 |
"""Train a KNN classifier on the student's 23 labeled polygons (using the
|
| 114 |
real spectral bands), then evaluate it only at the Ground Truth pixel
|
|
|
|
| 139 |
# with a known Ground Truth. We can't evaluate accuracy outside GT locations
|
| 140 |
# (no known answer there) but we can still show what the model predicted.
|
| 141 |
H, W, n_bands = bands.shape
|
| 142 |
+
pred_full = predict_full_image(clf, scaler.transform(bands.reshape(-1, n_bands)), H, W)
|
| 143 |
|
| 144 |
return model_oa, model_matrix, pred_gt, y_gt, pred_full
|
| 145 |
|
|
|
|
| 170 |
model_oa = model_matrix.diagonal().sum() / model_matrix.sum()
|
| 171 |
|
| 172 |
H, W, n_bands = bands.shape
|
| 173 |
+
pred_full = predict_full_image(clf, scaler.transform(bands.reshape(-1, n_bands)), H, W)
|
| 174 |
|
| 175 |
return model_oa, model_matrix, pred_gt, y_gt, pred_full
|
| 176 |
|
|
|
|
| 200 |
model_oa = model_matrix.diagonal().sum() / model_matrix.sum()
|
| 201 |
|
| 202 |
H, W, n_bands = bands.shape
|
| 203 |
+
pred_full = predict_full_image(clf, bands.reshape(-1, n_bands), H, W)
|
| 204 |
|
| 205 |
return model_oa, model_matrix, pred_gt, y_gt, pred_full
|
| 206 |
|
|
|
|
| 704 |
gr.update(value=table, visible=True),
|
| 705 |
gr.update(visible=True),
|
| 706 |
gr.update(value=student_fig),
|
| 707 |
+
gr.update(selected="tab_results"),
|
| 708 |
)
|
| 709 |
|
| 710 |
def on_train(current_labels, k):
|
|
|
|
| 897 |
results_tbl,
|
| 898 |
row_plots_student,
|
| 899 |
student_matrix_plot,
|
| 900 |
+
tabs,
|
| 901 |
],
|
| 902 |
)
|
| 903 |
|