Skip to content

Add object detection, tracking, mobile notifications, and search to any security camera.

revert

rorycleare5ffc6a88b69
diff --git a/clearcam.py b/clearcam.py
index 4b2ce2b..d785290 100644
--- a/clearcam.py
+++ b/clearcam.py
@@ -634,7 +634,7 @@ def point_not_in_polygon(coords, poly):
return True
def run_encode_text(return_q, searcher, text):
- res = searcher._encode_text(text)
+ res = searcher._encode_text(text, realize=True)
return_q.put(res)
return res
@@ -1291,7 +1291,7 @@ class ThreadedHTTPServer(ThreadingMixIn, HTTPServer):
for k, v in alerts.items():
if time.time() - v.last_det < 60 or not v.is_active(): continue
if v.desc is not None and hasattr(v, "desc_emb") and v.desc_emb is not None:
- similarity = (v.desc_emb.numpy() @ embedding.T).item()
+ similarity = (v.desc_emb @ embedding.T).item()
print("sim =",similarity,v.desc,path)
if similarity > v.threshold:
send_notif(userID, f"Event Detected ({name}: {v.desc})")
@@ -1454,3 +1454,4 @@ if __name__ == "__main__":
hls_streamer.stop()
cam.release()
server.shutdown()
+
diff --git a/clip_search.py b/clip_search.py
index 51c78e7..d1ba6a2 100644
--- a/clip_search.py
+++ b/clip_search.py
@@ -124,13 +124,15 @@ class CLIPSearch:
print(f"\nTotal images loaded: {total_loaded}")
- def _encode_text(self, query):
+ def _encode_text(self, query, realize=False):
tokens = [49406]
tokens += self.tokenizer.encode(query)
tokens.append(49407)
if len(tokens) < 77: tokens += [0] * (77 - len(tokens))
tokens = Tensor([tokens])
- return encode_text(self.model, tokens)
+ text_emb = encode_text(self.model, tokens)
+ if realize: return text_emb.numpy()
+ return text_emb
def search(self, query=None, top_k=10, cam_name=None, timestamp=None, text_embedding=None):
if not self.image_embeddings:
@@ -208,14 +210,3 @@ def encode_text(model, text):
x = x[0][argmax]
x = x @ model.text_projection
return x / (x * x).sum(axis=-1, keepdim=True).sqrt()
-
-'''
-if __name__ == "__main__":
- searcher = CLIPSearch()
- query = input("\nEnter search query (or 'quit' to exit): ").strip()
- results = searcher.search(query, top_k=10)
-
- print(f"\nTop results for '{query}' (best per object):")
- for i, (path, score) in enumerate(results, 1):
- print(f"{i}. Score: {score:.3f} - \"{path}\"")
-'''