chore : cctv 수정

chore : cctv 수정

.gitignore

@@ -12,4 +12,6 @@ aiServer/crowdhuman_yolov5m.pt
 ai/.env
 frontend/.env 
 server/src/main/java/com/capstone/server/dataInitializer/CCTVInitializer.java
-
+/ai/TextReID/datasets/imgs
+ai/imageSearch/results.json
+ai/imageSearch/search_result_grid.jpg

ai/TextReID/.gitignore

@@ -4,7 +4,6 @@
 *.npy
 
 output/
-datasets/
 pretrained/
 __pycache__/
 condor_log/

ai/TextReID/README.md

@@ -20,8 +20,9 @@ python train_net.py \
 python3 test_net.py \
 --config-file configs/cuhkpedes/moco_gru_cliprn50_ls_bs128_2048.yaml \
 --checkpoint-file output/cuhkpedes/moco_gru_cliprn50_ls_bs128_2048/best.pth \
---data-dir datasets/cuhkpedes \
---query "a man wearing a white black top and white pants"
+--data-dir /home/jongbin/Desktop/yolo/58 \
+--query "a boy has short hair. he wearing a white short sleeve and a khaki a long pants" 
+
 # 코드 안에서 QUERY를 받게 바꿨음
 ```
 

ai/TextReID/lib/data/datasets/cuhkpedes.py

@@ -0,0 +1,73 @@
+import json
+import os
+
+import torch
+from PIL import Image
+
+from lib.utils.caption import Caption
+
+
+class CUHKPEDESDataset(torch.utils.data.Dataset):
+    def __init__(
+        self,
+        root,
+        ann_file,
+        use_onehot=True,
+        max_length=100,
+        transforms=None,
+    ):
+        self.root = root
+        self.use_onehot = use_onehot
+        self.max_length = max_length
+        self.transforms = transforms
+
+        self.img_dir = os.path.join(self.root, "imgs")
+        # self.img_dir = self.root
+
+        print("loading annotations into memory...")
+        dataset = json.load(open(ann_file, "r"))
+        self.dataset = dataset["annotations"]
+
+    def __getitem__(self, index):
+        """
+        Args:
+            index(int): Index
+        Returns:
+            tuple: (images, labels, captions)
+        """
+        data = self.dataset[index]
+
+        img_path = data["file_path"]
+        img = Image.open(os.path.join(self.img_dir, img_path)).convert("RGB")
+
+        if self.use_onehot:
+            caption = data["onehot"]
+            caption = torch.tensor(caption)
+            caption = Caption([caption], max_length=self.max_length, padded=False)
+        else:
+            caption = data["sentence"]
+            caption = Caption(caption)
+
+        caption.add_field("img_path", img_path)
+
+        label = int(data["id"])
+
+        # Convert label to string representation and then to tensor of ASCII values
+        label_str = str(label)
+        label_tensor = torch.tensor([ord(char) for char in label_str])
+        caption.add_field("id", label_tensor)
+
+        if self.transforms is not None:
+            img = self.transforms(img)
+
+        query = data["sentence"]
+
+        return img, caption, index, query
+
+    def __len__(self):
+        return len(self.dataset)
+
+    def get_id_info(self, index):
+        image_id = self.dataset[index]["image_id"]
+        pid = self.dataset[index]["id"]
+        return image_id, pid

ai/TextReID/lib/data/metrics/evaluation.py

@@ -85,7 +85,7 @@ def evaluation(
     save_data=True,
     rerank=True,
     search_num=0,
-    save_folder = "./output/output.json"
+    save_folder="./output/output.json"
 ):
     logger = logging.getLogger("PersonSearch.inference")
     data_dir = os.path.join(output_folder, "inference_data.npz")
@@ -103,20 +103,26 @@ def evaluation(
         image_ids, pids = [], []
         image_global, text_global = [], []
 
-        # FIXME: need optimization
         for idx, prediction in predictions.items():
             image_id, pid = dataset.get_id_info(idx)
             image_ids.append(image_id)
             pids.append(pid)
             image_global.append(prediction[0])
             if len(prediction) == 2:
-                # text query를 하나만 넣었으므로, text emgedding은 배치의 제일 처음 이미지에만 들어감
-                # 왜냐하면 유사도 검사 시 배치 별로 검사를 했으니까
                 text_global.append(prediction[1])
 
-        pids = list(map(int, pids))
-        image_pid = torch.tensor(pids)
-        text_pid = torch.tensor(pids)
+        # Convert pids to string representation and then to tensor of ASCII values
+        pids = list(map(str, pids))
+        pids = [torch.tensor([ord(char) for char in pid]) for pid in pids]
+
+        # Find the maximum length of pids
+        max_length = max(len(pid) for pid in pids)
+
+        # Pad all pids to the same length
+        pids = [F.pad(pid, (0, max_length - len(pid)), value=0) for pid in pids]
+
+        image_pid = torch.stack(pids)
+        text_pid = torch.stack(pids)
         image_global = torch.stack(image_global, dim=0)
         text_global = torch.stack(text_global, dim=0)
 
@@ -129,21 +135,16 @@ def evaluation(
 
         similarity = torch.matmul(text_global, image_global.t())
 
-        # top 10 results 반환
         sorted_indices = torch.argsort(similarity[0], descending=True)
         sorted_values = similarity[0][sorted_indices]
         top_k = cfg.TEST.TOP_K
-        write = [cap[0]] # 저장할 output
+        write = [cap[0]]
         for index, value in zip(sorted_indices[:top_k], sorted_values[:top_k]):
-            # image_id, pid = dataset.get_id_info(idx)
             img, caption, idx, query = dataset.__getitem__(index)
             img_path = caption.get_field("img_path")
-            # print(f"Index: {index}, Similarity: {value}, pid: {pid}")
             dict = {"img_path": img_path, "Similarity": value.item()}
             write.append(dict)
 
-
-        #저장 폴더 생성
         if not os.path.exists(os.path.dirname(save_folder)):
             os.makedirs(os.path.dirname(save_folder), exist_ok=True)
 

ai/TextReID/test_net.py

@@ -80,7 +80,7 @@ def main():
     )
     parser.add_argument(
         "--top-k",
-        default=10,
+        default=400,
         type=int,
     )
 

ai/app.py

@@ -18,7 +18,7 @@
 sys.path.append(str(Path(__file__).parent)+"/yolov8")
 sys.path.append(str(Path(__file__).parent)+"/TextReID")
 sys.path.append(str(Path(__file__).parent)+"/imageSearch")
-
+import shutil
 from TextReID.test_net import findByText 
 from yolov5_crowdhuman.detect import run_detection
 from yolov8.run import run_Yolo
@@ -92,20 +92,35 @@ async def firstDetection(input :TotalInput):
         raise HTTPException(status_code=400, detail="cctv is undefined")
     yolo_save_path = f"/home/jongbin/Desktop/yolo/{input.searchId}" #경로는 각자 환경에 맞게 조장하시오
     run_Yolo(input.cctvId,yolo_save_path,input.startTime) #todo start time 따라 input다르게 만들기
+    # run_Yolo(input.cctvId,yolo_save_path,"2024-05-24T01:01:01") #todo start time 따라 input다르게 만들기
     result_dir = await runTextReID(input, yolo_save_path) #text-re-id돌리고 결과 json파일 받아오기
+    with open(result_dir, 'r') as json_file:
+        data = json.load(json_file)
+    image_paths = [entry['img_path'] for entry in data if 'img_path' in entry]
+    destination_folder = f'/home/jongbin/Desktop/firstResult/{input.searchId}'
+    os.makedirs(destination_folder, exist_ok=True)
+    # 이미지 복사
+    for img_path in image_paths:
+        if os.path.exists(img_path):
+            shutil.copy(img_path, destination_folder)
 
     result_json_dir = await uploadS3(result_dir,input.missingPeopleId, input.searchId, input.step) #json파일로 결과들 s3업로드하고 서버로 보낼 데이터 모음 json받아오기
     with open(result_json_dir, 'r') as file:
         result = json.load(file)
 
-    return DetectResult(searchId= input.searchId, missingPeopleId= input.missingPeopleId, data = result[1:])
+    # 이미지 경로 리스트
+
+
+    return DetectResult(searchId= input.searchId, missingPeopleId= input.missingPeopleId, data = result[1:16])
+
 #2차탐색
 @app.post("/second", response_model=DetectResult)
 async def secondDetection(input:SecondInput):
     print(input)
     img_download_url = "/home/jongbin/Desktop/imgDown"
-    data_path = f"/home/jongbin/Desktop/yolo/{input.firstSearchId}"
+    data_path = f"/home/jongbin/Desktop/firstResult/{input.firstSearchId}"
     result = []
+    check = set()
     for img_path in input.queryImagePath:
         #img_path는 s3주소 특정폴더에 다운로드하고 경로가져오기
         local_image_path = os.path.join(img_download_url, os.path.basename(img_path))
@@ -115,14 +130,19 @@ async def secondDetection(input:SecondInput):
         for output in data_to_save['output']:
             #해당 주소에 있는 이미지 s3업로드하고 이미지 주소 result에 넣기
             local_output_path = output['output_dir']
+            if local_output_path in check:
+                continue
+            check.add(local_output_path)
             new_file_name = f"{os.path.basename(local_output_path).split('.')[0]}"
             new_file_name = new_file_name.replace(' ', '-').replace(':', '').replace('/', '+')
             s3_key = f"missingPeopleId={input.missingPeopleId}/searchHistoryId={input.secondSearchId}/step=second/{new_file_name}"
             s3_url = upload_image_to_s3(local_output_path, s3_key)
-            a = {"img_path" : s3_url, "cctvId" : new_file_name.split('_')[0], "similarity" :output['score'] } #todo : 버그수정
+            a = {"img_path" : s3_url, "cctvId" : new_file_name.split('_')[0], "similarity" :(10000-output['score'])/10000 }
+            # a = {"img_path" : s3_url, "cctvId" : new_file_name.split('_')[0], "similarity" :output['score']}
             result.append(a)
 
-        result = sorted(result,key = lambda x:x["score"])
+        result = sorted(result,key = lambda x:x["similarity"])
+        print(result)
     return DetectResult(searchId= input.secondSearchId, missingPeopleId= 1, data = result)
 
 async def runTextReID(input : TotalInput, yolo_save_path:str):
@@ -132,7 +152,7 @@ async def runTextReID(input : TotalInput, yolo_save_path:str):
     home_path = os.path.expanduser("~")
     result_dir = os.path.join(home_path, "Desktop", "result", str(input.searchId) ,"output.json")
     # findByText(root_path, search_num=input.searchId, query = input.query, data_dir = yolo_save_path, save_folder = result_dir)
-    findByText(root_path, search_num=input.searchId, query = input.query, data_dir = yolo_save_path, save_folder = result_dir)
+    findByText(root_path, search_num=input.searchId, query = input.query, data_dir = yolo_save_path, save_folder = result_dir,result_num=200)
     return result_dir
 
 
@@ -146,7 +166,7 @@ async def uploadS3(json_file_path:str, missingPeopleId:int, searchId:int, step:s
     # 이미지 파일을 S3에 업로드하고 URL 업데이트
     errors = []
     # 리스트의 첫 번째 요소는 무시하고 나머지에서 작업 수행
-    for item in data[1:]:
+    for item in data[1:16]:
         img_path = item['img_path']
         similarity = item['Similarity']
         new_file_name = f"{os.path.basename(img_path).split('.')[0]}_{similarity}{os.path.splitext(img_path)[-1]}"
@@ -208,7 +228,13 @@ def upload_image_to_s3(local_path, s3_key):
     """
     로컬 경로의 이미지를 S3 버킷의 지정된 키에 업로드합니다.
     """
-    s3_client.upload_file(local_path, bucket_name, s3_key,ExtraArgs={
-        'ACL': 'public-read'  # 원하는 ACL 권한으로 설정
-    })
+    with open(local_path, 'rb') as img_file:
+                s3_client.upload_fileobj(
+                    Fileobj=img_file,
+                    Bucket=bucket_name,
+                    Key=s3_key,
+                    ExtraArgs={
+                    'ACL': 'public-read'  # 공개적으로 읽을 수 있도록 권한 설정
+            })
+
     return f"https://{bucket_name}.s3.amazonaws.com/{s3_key}"

ai/forTest.py

@@ -0,0 +1,30 @@
+import json
+import os
+import shutil
+
+# JSON 파일 경로
+json_file_path = '/home/jongbin/Documents/GitHub/capstone-2024-14/capstone-2024-14/ai/TextReID/output/output.json'
+
+# 이미지를 복사할 대상 폴더 경로
+destination_folder = '/home/jongbin/Desktop/copied_images'
+
+# 대상 폴더가 존재하지 않으면 생성
+if not os.path.exists(destination_folder):
+    os.makedirs(destination_folder)
+
+# JSON 파일 읽기
+with open(json_file_path, 'r') as json_file:
+    data = json.load(json_file)
+
+# 이미지 경로 리스트
+image_paths = [entry['img_path'] for entry in data if 'img_path' in entry]
+
+# 이미지 복사
+for img_path in image_paths:
+    if os.path.exists(img_path):
+        shutil.copy(img_path, destination_folder)
+        print(f"Copied: {img_path}")
+    else:
+        print(f"File not found: {img_path}")
+
+print("Image copying completed.")

ai/imageSearch/imgSearch.py

@@ -119,7 +119,7 @@ def save_results(query_image_path, scores, retrieved_examples, sorted_indices):
 
 # ===== 결과 저장
 scores, retrieved_examples = get_neighbors(args.query, args.topk)
-sorted_indices = np.argsort(scores)[::-1]
+sorted_indices = np.argsort(scores)[::]
 
 images = [Image.open(args.query)]
 images.extend([Image.open(retrieved_examples["image_path"][i]) for i in sorted_indices])

ai/imageSearch/imgSearch_server.py

@@ -61,13 +61,13 @@ def run_Image_to_Image(data_path:str,topk:int,query:str):
     dataset.add_faiss_index(column='embeddings')
     # ===== 결과 저장
     scores, retrieved_examples = get_neighbors(query, dataset, topk)
-
+    sorted_indices = np.argsort(scores)[::]
 
     sec = time.time() - start_time
     times = str(datetime.timedelta(seconds=sec))
     short = times.split(".")[0]
     print(f"The 2nd search has ended. \nThe time required: {short} sec\n")
-    return save_results(query, scores, retrieved_examples)
+    return save_results(query, scores, retrieved_examples,sorted_indices)
 
 # ===== 사전학습 모델 세팅 및 특징 추출기 세팅
 def load_image_paths(data_dir):
@@ -96,12 +96,12 @@ def default(o):
     if isinstance(o, np.float32):
         return float(o)
 
-def save_results(query_image_path, scores, retrieved_examples):
+def save_results(query_image_path, scores, retrieved_examples,sorted_indices):
     data_to_save = {
         "query_dir": query_image_path,
         "output": [
             {"output_dir": retrieved_examples['image_path'][i], "score": float(scores[i])}
-            for i in range(len(scores))
+            for i in sorted_indices
         ]
     }
     return data_to_save

ai/imageSearch/readme.md

@@ -1,3 +1,5 @@
-python3 imgSearch.py --data_path /home/jongbin/Desktop/yolo/164 --topk 20 --query /home/jongbin/Desktop/yolo/164/1_2024-05-14_16-07-56_155.jpg
+python3 imgSearch.py --data_path /home/jongbin/Desktop/copied_images --topk 20 --query /home/jongbin/Desktop/copied_images/5_2024-05-14_16-26-33_133.jpg
+
+python3 imgSearch.py --data_path /home/jongbin/Desktop/yolo/23 --topk 40 --query /home/jongbin/Desktop/copied_images/1_2024-05-14_16-07-56_387.jpg
 
 /home/jongbin/Desktop/yolo/164

ai/yolov8/run.py

@@ -39,6 +39,8 @@ def run_Yolo(cctvIds,save_path,startTime):
     annotations = []
     for cctv_id in cctvIds:
         source_path = f'/home/jongbin/Desktop/cctv/{cctv_id.id}/{startTime.split("T")[0]}' #todo 세부적인 시 분 초 시간까지 골라서 돌리기 기능 추가
+        if not os.path.exists(source_path):
+            continue  # source_path가 존재하지 않으면 다음 반복으로 넘어감
         for filename in os.listdir(source_path):
             source = os.path.join(source_path, filename)
             start_time = extract_video_info(filename)

server/src/main/java/com/capstone/server/controller/CCTVController.java

@@ -10,10 +10,13 @@
 
 import com.capstone.server.dto.CCTVDto;
 import com.capstone.server.model.CCTVEntity;
-import com.capstone.server.model.CCTVEntity;
 import com.capstone.server.repository.CCTVRepository;
 import com.capstone.server.response.SuccessResponse;
 import com.capstone.server.service.CCTVService;
+import jakarta.transaction.Transactional;
+import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.http.ResponseEntity;
+import org.springframework.web.bind.annotation.*;
 
 import jakarta.transaction.Transactional;
 
@@ -30,10 +33,10 @@ public class CCTVController {
 
     @GetMapping
     public ResponseEntity<?> getAllCCTV(
-        @RequestParam(required = true, value = "longitude") double longitude,
-        @RequestParam(required = true, value = "latitude") double latitude,
-        @RequestParam(required = false, defaultValue = "1000", value = "distance") double distance) {
-        
+            @RequestParam(required = true, value = "longitude") double longitude,
+            @RequestParam(required = true, value = "latitude") double latitude,
+            @RequestParam(required = false, defaultValue = "1000", value = "distance") double distance) {
+
         List<CCTVDto> CCTVList = cctvService.findCCTVsNearbyLocationWithinDistance(longitude, latitude);
         return ResponseEntity.ok().body(CCTVList);
     }