yixiong/FoodClassifier
3
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

代码回滚：增加未知类别负样本之后，模型不收敛，训练效果特别差，现在进行回滚。回滚之后效果非常不错。

Browse Source
This commit is contained in:
zhangpu
2025-10-29 16:40:05 +08:00
parent 1c8ceef5b0
commit 07bae61a10
4 changed files with 43 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files
classifier/embedding_food_classifier_app.py
+14 -9
View File
@@ -65,12 +65,14 @@ class EmbeddingFoodClassifierApp:
BASE_DIR = os.path.dirname(os.path.abspath(__file__)) BASE_DIR = os.path.dirname(os.path.abspath(__file__))
# model_path = "../model/embedding_20251011_133653/best_embedding_model.pth" # model_path = "../model/embedding_20251011_133653/best_embedding_model.pth"
# model_path = os.path.join(BASE_DIR, "../model/embedding_20251011_133653/best_embedding_model.pth") # model_path = os.path.join(BASE_DIR, "../model/embedding_20251011_133653/best_embedding_model.pth")
model_path = os.path.join(BASE_DIR, "../model/WholeIngredientRecognition/embedding_20251021_085915/best_embedding_model.pth") model_path = os.path.join(BASE_DIR, "../model/ProcessedIngredientRecognition/embedding_20251029_100204/best_embedding_model.pth")
# model_path = os.path.join(BASE_DIR, "../model/WholeIngredientRecognition/embedding_20251024_091151/best_embedding_model.pth")
# model_path = os.path.join(BASE_DIR, "../model/DishClassification/embedding_20251022_093635/best_embedding_model.pth") # model_path = os.path.join(BASE_DIR, "../model/DishClassification/embedding_20251022_093635/best_embedding_model.pth")
# FAISS索引目录 # FAISS索引目录
# index_dir = "../faiss_vector_db/faiss_index" # index_dir = "../faiss_vector_db/faiss_index"
# index_dir = os.path.join(BASE_DIR, "../faiss_vector_db/faiss_index") # index_dir = os.path.join(BASE_DIR, "../faiss_vector_db/faiss_index")
index_dir = os.path.join(BASE_DIR, "../faiss_vector_db/ProcessedIngredientRecognition/faiss_index")
index_dir = os.path.join(BASE_DIR, "../faiss_vector_db/WholeIngredientRecognition/faiss_index") index_dir = os.path.join(BASE_DIR, "../faiss_vector_db/WholeIngredientRecognition/faiss_index")
# index_dir = os.path.join(BASE_DIR, "../faiss_vector_db/DishClassification/faiss_index") # index_dir = os.path.join(BASE_DIR, "../faiss_vector_db/DishClassification/faiss_index")
@@ -1087,12 +1089,13 @@ class EmbeddingFoodClassifierApp:
"""识别所有图片""" """识别所有图片"""
try: try:
self.current_results.clear() self.current_results.clear()
score_list = []
for i, img_info in enumerate(self.uploaded_images): for i, img_info in enumerate(self.uploaded_images):
# 使用embedding相似度识别 # 使用embedding相似度识别
if self.model is not None and self.faiss_index is not None: if self.model is not None and self.faiss_index is not None:
# 使用真实的embedding模型和FAISS索引 # 使用真实的embedding模型和FAISS索引
predicted_class, confidence, similar_images = self.predict_with_embedding(img_info['image']) predicted_class, confidence, similar_images,score = self.predict_with_embedding(img_info['image'])
score_list.append(score)
else: else:
# 模拟预测结果 # 模拟预测结果
predicted_class = np.random.choice(self.class_names) predicted_class = np.random.choice(self.class_names)
@@ -1122,7 +1125,7 @@ class EmbeddingFoodClassifierApp:
# 更新UI（在主线程中） # 更新UI（在主线程中）
self.root.after(0, self.update_progress, i + 1, len(self.uploaded_images)) self.root.after(0, self.update_progress, i + 1, len(self.uploaded_images))
print('socre_list',score_list)
# 识别完成，更新UI # 识别完成，更新UI
self.root.after(0, self.recognition_completed) self.root.after(0, self.recognition_completed)
@@ -1144,9 +1147,11 @@ class EmbeddingFoodClassifierApp:
# 在FAISS索引中搜索最相似的k张图片 # 在FAISS索引中搜索最相似的k张图片
scores, indices = self.faiss_index.search(query_embedding, k) scores, indices = self.faiss_index.search(query_embedding, k)
print("最相似的图片索引:", indices) # print("最相似的图片索引:", indices)
print("最相似的图片分数:", scores) # print("最相似的图片分数:", type(scores[0]))
# print("最相似的图片分数:", scores[0,0])
# print("最相似的图片分数:", type(scores[0,0]))
# 收集相似图片的类别 # 收集相似图片的类别
similar_classes = [] similar_classes = []
similar_images = [] similar_images = []
@@ -1179,12 +1184,12 @@ class EmbeddingFoodClassifierApp:
vote_ratio = class_counts[predicted_class] / len(similar_classes) vote_ratio = class_counts[predicted_class] / len(similar_classes)
confidence = max_score * vote_ratio confidence = max_score * vote_ratio
return predicted_class, confidence, similar_images return predicted_class, confidence, similar_images,round(scores[0,0],3)
else: else:
# 如果没有找到相似图片，随机选择一个类别 # 如果没有找到相似图片，随机选择一个类别
predicted_class = np.random.choice(self.class_names) predicted_class = np.random.choice(self.class_names)
confidence = 0.1 confidence = 0.1
return predicted_class, confidence, [] return predicted_class, confidence, [],scores[0,0]
except Exception as e: except Exception as e:
print(f"Embedding预测出错: {e}") print(f"Embedding预测出错: {e}")
faiss_vector_db/build_faiss_index.py
+7 -4
View File
@@ -480,12 +480,15 @@ def main():
"""主函数""" """主函数"""
# 配置参数 # 配置参数
# MODEL_PATH = "../model/embedding_20251011_133653/best_embedding_model.pth" # MODEL_PATH = "../model/embedding_20251011_133653/best_embedding_model.pth"
# MODEL_PATH = "../model/WholeIngredientRecognition/embedding_20251021_085915/best_embedding_model.pth" MODEL_PATH = "../model/ProcessedIngredientRecognition/embedding_20251029_100204/best_embedding_model.pth"
MODEL_PATH = "../model/DishClassification/embedding_20251022_093635/best_embedding_model.pth" # MODEL_PATH = "../model/WholeIngredientRecognition/embedding_20251024_091151/best_embedding_model.pth"
# MODEL_PATH = "../model/DishClassification/embedding_20251022_093635/best_embedding_model.pth"
TRAIN_DIR = "../dataset/ProcessedIngredientRecognition/train"
# TRAIN_DIR = "../dataset/WholeIngredientRecognition/train" # TRAIN_DIR = "../dataset/WholeIngredientRecognition/train"
TRAIN_DIR = "../dataset/DishClassification/train" # TRAIN_DIR = "../dataset/DishClassification/train"
OUTPUT_DIR = "ProcessedIngredientRecognition/faiss_index"
# OUTPUT_DIR = "WholeIngredientRecognition/faiss_index" # OUTPUT_DIR = "WholeIngredientRecognition/faiss_index"
OUTPUT_DIR = "DishClassification/faiss_index" # OUTPUT_DIR = "DishClassification/faiss_index"
BATCH_SIZE = 16 BATCH_SIZE = 16
INDEX_TYPE = 'flat' # 'flat', 'ivf', 'hnsw' INDEX_TYPE = 'flat' # 'flat', 'ivf', 'hnsw'
EMBEDDING_DIM = 512 EMBEDDING_DIM = 512
faiss_vector_db/visualize_embeddings.py
+4 -2
View File
@@ -162,10 +162,12 @@ def plot_2d(Z: np.ndarray, y: np.ndarray, title: str, out_path: Optional[str] =
def main(): def main():
parser = argparse.ArgumentParser(description="可视化高维 embedding 并进行坍塌诊断") parser = argparse.ArgumentParser(description="可视化高维 embedding 并进行坍塌诊断")
parser.add_argument("--embeddings", type=str, default=os.path.join("DishClassification/faiss_index", "embeddings.json"), help="embeddings.json 路径") # parser.add_argument("--embeddings", type=str, default=os.path.join("DishClassification/faiss_index", "embeddings.json"), help="embeddings.json 路径")
# parser.add_argument("--embeddings", type=str, default=os.path.join("WholeIngredientRecognition/faiss_index", "embeddings.json"), help="embeddings.json 路径") # parser.add_argument("--embeddings", type=str, default=os.path.join("WholeIngredientRecognition/faiss_index", "embeddings.json"), help="embeddings.json 路径")
parser.add_argument("--labels", type=str, default=os.path.join("DishClassification/faiss_index", "labels.json"), help="labels.json 路径") parser.add_argument("--embeddings", type=str, default=os.path.join("ProcessedIngredientRecognition/faiss_index", "embeddings.json"), help="embeddings.json 路径")
# parser.add_argument("--labels", type=str, default=os.path.join("DishClassification/faiss_index", "labels.json"), help="labels.json 路径")
# parser.add_argument("--labels", type=str, default=os.path.join("WholeIngredientRecognition/faiss_index", "labels.json"), help="labels.json 路径") # parser.add_argument("--labels", type=str, default=os.path.join("WholeIngredientRecognition/faiss_index", "labels.json"), help="labels.json 路径")
parser.add_argument("--labels", type=str, default=os.path.join("ProcessedIngredientRecognition/faiss_index", "labels.json"), help="labels.json 路径")
parser.add_argument("--method", type=str, default="pca", choices=["pca", "tsne", "umap"], help="降维方法") parser.add_argument("--method", type=str, default="pca", choices=["pca", "tsne", "umap"], help="降维方法")
parser.add_argument("--seed", type=int, default=42) parser.add_argument("--seed", type=int, default=42)
parser.add_argument("--max_points", type=int, default=None, help="抽样上限，避免t-SNE/UMAP过慢；None为全量") parser.add_argument("--max_points", type=int, default=None, help="抽样上限，避免t-SNE/UMAP过慢；None为全量")
train/train_embedding.py
+18 -15
View File
@@ -73,9 +73,11 @@ TASKS = {
embedding_dim=512, embedding_dim=512,
batch_size=16, batch_size=16,
lr=1e-3, lr=1e-3,
triplet_margin=0.25, # triplet_margin=0.25,
center_loss_weight=0.1, triplet_margin=0.5,
aug_strength="shape", # center_loss_weight=0.5,
center_loss_weight=20,
aug_strength="medium",
), ),
} }
@@ -142,11 +144,11 @@ class TripletDataset(Dataset):
三元组数据集，用于三元组损失训练 三元组数据集，用于三元组损失训练
每个样本包含：锚点(anchor)、正样本(positive)、负样本(negative) 每个样本包含：锚点(anchor)、正样本(positive)、负样本(negative)
""" """
def __init__(self, dataset_path: str, transform=None, samples_per_class: int = 100): def __init__(self, dataset_path: str, transform=None, samples_per_class: int = 100):
""" """
初始化三元组数据集 初始化三元组数据集
Args: Args:
dataset_path: 数据集路径 dataset_path: 数据集路径
transform: 数据变换 transform: 数据变换
@@ -155,13 +157,13 @@ class TripletDataset(Dataset):
self.dataset_path = dataset_path self.dataset_path = dataset_path
self.transform = transform self.transform = transform
self.samples_per_class = samples_per_class self.samples_per_class = samples_per_class
# 加载数据集 # 加载数据集
self.class_to_idx = {} self.class_to_idx = {}
self.idx_to_class = {} self.idx_to_class = {}
self.samples_by_class = defaultdict(list) self.samples_by_class = defaultdict(list)
self.all_samples = [] self.all_samples = []
self._load_dataset() self._load_dataset()
def _load_dataset(self): def _load_dataset(self):
@@ -191,24 +193,24 @@ class TripletDataset(Dataset):
logger.info(f"加载数据集完成:") logger.info(f"加载数据集完成:")
for class_name, class_idx in self.class_to_idx.items(): for class_name, class_idx in self.class_to_idx.items():
logger.info(f" {class_name}: {len(self.samples_by_class[class_idx])} 张图片") logger.info(f" {class_name}: {len(self.samples_by_class[class_idx])} 张图片")
def __len__(self): def __len__(self):
return len(self.all_samples) return len(self.all_samples)
def __getitem__(self, idx): def __getitem__(self, idx):
""" """
获取三元组样本 获取三元组样本
Returns: Returns:
tuple: (anchor, positive, negative, anchor_label) tuple: (anchor, positive, negative, anchor_label)
""" """
# 获取锚点样本 # 获取锚点样本
anchor_path, anchor_label = self.all_samples[idx] anchor_path, anchor_label = self.all_samples[idx]
anchor_img = self._load_image(anchor_path) anchor_img = self._load_image(anchor_path)
# 获取正样本（同类别的不同图片） # 获取正样本（同类别的不同图片）
# 需要从自己所在类别中先把自己给排除掉 # 需要从自己所在类别中先把自己给排除掉
positive_candidates = [path for path in self.samples_by_class[anchor_label] positive_candidates = [path for path in self.samples_by_class[anchor_label]
if path != anchor_path] if path != anchor_path]
if positive_candidates: if positive_candidates:
# 从这里可以看出来是随机选的 # 从这里可以看出来是随机选的
@@ -216,7 +218,7 @@ class TripletDataset(Dataset):
else: else:
positive_path = anchor_path # 如果只有一张图片，使用自己作为正样本 positive_path = anchor_path # 如果只有一张图片，使用自己作为正样本
positive_img = self._load_image(positive_path) positive_img = self._load_image(positive_path)
# 获取负样本（不同类别的图片） # 获取负样本（不同类别的图片）
# 获得其它类别 # 获得其它类别
negative_classes = [cls for cls in self.samples_by_class.keys() if cls != anchor_label] negative_classes = [cls for cls in self.samples_by_class.keys() if cls != anchor_label]
@@ -225,7 +227,7 @@ class TripletDataset(Dataset):
# 随便选一个路径 # 随便选一个路径
negative_path = random.choice(self.samples_by_class[negative_class]) negative_path = random.choice(self.samples_by_class[negative_class])
negative_img = self._load_image(negative_path) negative_img = self._load_image(negative_path)
return anchor_img, positive_img, negative_img, anchor_label return anchor_img, positive_img, negative_img, anchor_label
def _load_image(self, image_path: str): def _load_image(self, image_path: str):
@@ -818,5 +820,6 @@ if __name__ == "__main__":
parser = argparse.ArgumentParser() parser = argparse.ArgumentParser()
parser.add_argument("task", choices=list(TASKS.keys()), nargs="?", default="dish") parser.add_argument("task", choices=list(TASKS.keys()), nargs="?", default="dish")
args = parser.parse_args() args = parser.parse_args()
main("processed_ingredient")
# main("whole_ingredient") # main("whole_ingredient")
main("dish") # main("dish")