From 0e8a1df59002033c8069ee0277cefd54b3f3e468 Mon Sep 17 00:00:00 2001 From: zhangpu <1250681871@qq.com> Date: Wed, 26 Nov 2025 14:50:00 +0800 Subject: [PATCH] =?UTF-8?q?=E4=B8=8D=E8=A6=81=E7=B1=BB=E5=88=AB=E5=8A=A0?= =?UTF-8?q?=E6=9D=83?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- classifier/embedding_food_classifier_app.py | 7 ++++--- faiss_vector_db/build_faiss_index.py | 12 ++++++------ faiss_vector_db/visualize_embeddings.py | 8 ++++---- toAndroid/toAndroidEmbedding.py | 10 +++++----- train/grid_search_cosface.py | 12 +++++++++--- 5 files changed, 28 insertions(+), 21 deletions(-) diff --git a/classifier/embedding_food_classifier_app.py b/classifier/embedding_food_classifier_app.py index 96a4bad..06c4396 100644 --- a/classifier/embedding_food_classifier_app.py +++ b/classifier/embedding_food_classifier_app.py @@ -67,14 +67,15 @@ class EmbeddingFoodClassifierApp: # model_path = os.path.join(BASE_DIR, "../model/embedding_20251011_133653/best_embedding_model.pth") # model_path = os.path.join(BASE_DIR, "../model/ProcessedIngredientRecognition/embedding_20251029_170904/best_embedding_model.pth") # model_path = os.path.join(BASE_DIR, "../model/WholeIngredientRecognition/cosface_20251106_134718/best_cosface_model.pth") - model_path = os.path.join(BASE_DIR, "../model/DishClassification/cosface_20251111_153649/best_cosface_model.pth") + model_path = os.path.join(BASE_DIR, "../model/WholeIngredientRecognition/grid_search_20251120_205839/model_s68.0_m0.38.pth") + # model_path = os.path.join(BASE_DIR, "../model/DishClassification/cosface_20251111_153649/best_cosface_model.pth") # FAISS索引目录 # 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/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/DishClassification/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") if os.path.exists(model_path) and os.path.exists(index_dir): # 1. 加载embedding模型 diff --git a/faiss_vector_db/build_faiss_index.py b/faiss_vector_db/build_faiss_index.py index dced3f0..7c2ba7f 100644 --- a/faiss_vector_db/build_faiss_index.py +++ b/faiss_vector_db/build_faiss_index.py @@ -498,16 +498,16 @@ def main(): # MODEL_PATH = "../model/embedding_20251011_133653/best_embedding_model.pth" # MODEL_PATH = "../model/ProcessedIngredientRecognition/embedding_20251103_172012/best_embedding_model.pth" # MODEL_PATH = "../model/WholeIngredientRecognition/cosface_20251113_160103/best_cosface_model.pth" - MODEL_PATH = "../model/WholeIngredientRecognition/grid_search_20251113_095659/model_s68.0_m0.4.pth" - # MODEL_PATH = "../model/DishClassification/cosface_20251105_200551/best_embedding_model.pth" + # MODEL_PATH = "../model/WholeIngredientRecognition/grid_search_20251120_205839/model_s68.0_m0.38.pth" + MODEL_PATH = "../model/DishClassification/grid_search_20251121_102723/model_s56.0_m0.4.pth" # MODEL_PATH = "../model/DishClassification/cosface_20251111_153649/best_cosface_model.pth" # TRAIN_DIR = "../dataset/ProcessedIngredientRecognition/train" - TRAIN_DIR = "../dataset/WholeIngredientRecognition/train" - # TRAIN_DIR = "../dataset/DishClassification/train" + # TRAIN_DIR = "../dataset/WholeIngredientRecognition/train" + TRAIN_DIR = "../dataset/DishClassification/train" # OUTPUT_DIR = "ProcessedIngredientRecognition/faiss_index" - OUTPUT_DIR = "WholeIngredientRecognition/faiss_index" - # OUTPUT_DIR = "DishClassification/faiss_index" + # OUTPUT_DIR = "WholeIngredientRecognition/faiss_index" + OUTPUT_DIR = "DishClassification/faiss_index" BATCH_SIZE = 16 INDEX_TYPE = 'flat' # 'flat', 'ivf', 'hnsw' EMBEDDING_DIM = 512 diff --git a/faiss_vector_db/visualize_embeddings.py b/faiss_vector_db/visualize_embeddings.py index ae6eb9a..aacfecc 100644 --- a/faiss_vector_db/visualize_embeddings.py +++ b/faiss_vector_db/visualize_embeddings.py @@ -162,11 +162,11 @@ def plot_2d(Z: np.ndarray, y: np.ndarray, title: str, out_path: Optional[str] = def main(): 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("WholeIngredientRecognition/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("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("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("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("--seed", type=int, default=42) diff --git a/toAndroid/toAndroidEmbedding.py b/toAndroid/toAndroidEmbedding.py index 4b6088d..fc7db4f 100644 --- a/toAndroid/toAndroidEmbedding.py +++ b/toAndroid/toAndroidEmbedding.py @@ -17,9 +17,9 @@ def main(): # 1. 加载训练好的embedding模型权重 # base_model = create_resnet50_embedding(embedding_dim=512, pretrained=True) base_model = create_mobile_resnet50_embedding(embedding_dim=512, pretrained=True) - # model_path = "../model/DishClassification/cosface_20251110_144822/best_cosface_model.pth" + model_path = "../model/DishClassification/grid_search_20251121_102723/model_s56.0_m0.4.pth" # model_path = "../model/WholeIngredientRecognition/cosface_20251106_134718/best_cosface_model.pth" - model_path = "../model/WholeIngredientRecognition/grid_search_20251113_095659/model_s68.0_m0.4.pth" + # model_path = "../model/WholeIngredientRecognition/grid_search_20251120_205839/model_s68.0_m0.38.pth" # model_path = "../model/ProcessedIngredientRecognition/embedding_20251029_173607/best_embedding_model.pth" if not os.path.exists(model_path): @@ -92,9 +92,9 @@ def main(): # 使用trace方法转换 traced_model = torch.jit.trace(mobile_wrapper, single_input) - # 保存模型 - # output_path = "../model/DishClassification/cosface_20251110_144822/best_embedding_model_mobile.pt" - output_path = "../model/WholeIngredientRecognition/grid_search_20251113_095659/best_embedding_model_mobile.pt" + # 保存模型,一定不要乱改，防止模型覆盖!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! + output_path = "../model/DishClassification/grid_search_20251121_102723/best_embedding_model_mobile.pt" + # output_path = "../model/WholeIngredientRecognition/grid_search_20251120_205839/best_embedding_model_mobile.pt" # output_path = "../model/WholeIngredientRecognition/cosface_20251106_134718/best_embedding_model_mobile.pt" # output_path = "../model/ProcessedIngredientRecognition/embedding_20251029_173607/best_embedding_model_mobile.pt" traced_model.save(output_path) diff --git a/train/grid_search_cosface.py b/train/grid_search_cosface.py index aa39199..2ab75c8 100644 --- a/train/grid_search_cosface.py +++ b/train/grid_search_cosface.py @@ -41,10 +41,16 @@ logging.basicConfig( logger = logging.getLogger(__name__) # ==================== 网格搜索空间定义 ==================== +""" +'dish': { + 's': [56.0, 60.0, 64.0, 68.0], # scale参数 + 'm': [0.32, 0.35, 0.38, 0.40], # margin参数 + }, +""" GRID_PARAMS = { 'dish': { 's': [56.0, 60.0, 64.0, 68.0], # scale参数 - 'm': [0.32, 0.35, 0.38, 0.40], # margin参数 + 'm': [0.32, 0.35, 0.38, 0.40,0.45,0.50], # margin参数 }, 'whole_ingredient': { 's': [56.0, 60.0, 64.0, 68.0], @@ -354,8 +360,8 @@ def grid_search_main( if __name__ == '__main__': import argparse parser = argparse.ArgumentParser(description='CosFace超参数网格搜索') - parser.add_argument('--task', choices=list(TASKS.keys()), default='dish', help='任务名称') - # parser.add_argument('--task', choices=list(TASKS.keys()), default='whole_ingredient', help='任务名称') + # parser.add_argument('--task', choices=list(TASKS.keys()), default='dish', help='任务名称') + parser.add_argument('--task', choices=list(TASKS.keys()), default='whole_ingredient', help='任务名称') parser.add_argument('--max_configs', type=int, default=None, help='最大配置数（用于测试）') parser.add_argument('--epochs', type=int, default=100, help='每个配置的最大训练轮数') parser.add_argument('--patience', type=int, default=10, help='早停容忍轮数')