Spaces:
Sleeping
Sleeping
File size: 3,887 Bytes
52dd1ca |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 |
# src/training/train_resnet_pt_svm.py
import os
import argparse
import json
import numpy as np
from sklearn.svm import LinearSVC
from sklearn.metrics import accuracy_score
import joblib
def load_features(features_dir: str):
x_train_path = os.path.join(features_dir, "X_train_resnet18.npy")
y_train_path = os.path.join(features_dir, "y_train.npy")
x_test_path = os.path.join(features_dir, "X_test_resnet18.npy")
y_test_path = os.path.join(features_dir, "y_test.npy")
assert os.path.exists(x_train_path), f"Missing: {x_train_path}"
assert os.path.exists(y_train_path), f"Missing: {y_train_path}"
assert os.path.exists(x_test_path), f"Missing: {x_test_path}"
assert os.path.exists(y_test_path), f"Missing: {y_test_path}"
X_train = np.load(x_train_path)
y_train = np.load(y_train_path)
X_test = np.load(x_test_path)
y_test = np.load(y_test_path)
return X_train, y_train, X_test, y_test
def main(
features_dir: str = "data/resnet18_features",
ckpt_path: str = "checkpoints/resnet_pt_svm_head.joblib",
labels_path: str = "configs/labels.json",
):
os.makedirs(os.path.dirname(ckpt_path), exist_ok=True)
print(f"[+] Loading ResNet18 features from {features_dir} ...")
X_train, y_train, X_test, y_test = load_features(features_dir)
print(f" X_train shape: {X_train.shape}, y_train shape: {y_train.shape}")
print(f" X_test shape : {X_test.shape}, y_test shape : {y_test.shape}")
num_features = X_train.shape[1]
print(f"[+] Feature dimension: {num_features}")
# Labels mapping for logging / sanity
if os.path.exists(labels_path):
with open(labels_path, "r") as f:
labels = json.load(f)
num_classes = len(labels)
print(f"[+] Loaded labels from {labels_path}, num_classes={num_classes}")
else:
print(f"[!] Warning: {labels_path} not found. Inference will need this later.")
labels = None
print("[+] Training Linear SVM on ResNet18 features ...")
svm = LinearSVC(
C=1.0,
penalty="l2",
loss="squared_hinge",
max_iter=5000, # give it some room
)
svm.fit(X_train, y_train)
print("[+] Evaluating ...")
y_pred_train = svm.predict(X_train)
y_pred_test = svm.predict(X_test)
train_acc = accuracy_score(y_train, y_pred_train)
test_acc = accuracy_score(y_test, y_pred_test)
print(f" Train accuracy: {train_acc:.4f}")
print(f" Test accuracy : {test_acc:.4f}")
print(f"[+] Saving ResNet PT + SVM head to {ckpt_path} ...")
payload = {
"model": svm,
"backbone": "resnet18_imagenet",
"feature_dim": int(num_features),
"labels_path": labels_path,
"train_acc": float(train_acc),
"test_acc": float(test_acc),
}
joblib.dump(payload, ckpt_path)
print("[+] Done training ResNet PT + SVM.")
def parse_args():
parser = argparse.ArgumentParser(
description="Train Linear SVM head on ResNet18 (pretrained) features."
)
parser.add_argument(
"--features-dir",
type=str,
default="data/resnet18_features",
help="Directory containing X_train_resnet18.npy etc.",
)
parser.add_argument(
"--ckpt-path",
type=str,
default="checkpoints/resnet_pt_svm_head.joblib",
help="Where to save SVM head checkpoint.",
)
parser.add_argument(
"--labels-path",
type=str,
default="configs/labels.json",
help="Path to labels mapping JSON.",
)
return parser.parse_args()
if __name__ == "__main__":
args = parse_args()
main(
features_dir=args.features_dir,
ckpt_path=args.ckpt_path,
labels_path=args.labels_path,
)
|