Post #3261477

For yesterdays rally at London, I built a crowd counting system from scratch on Apple Silicon 🍎 Trained a custom Swin-T + CNN hybrid on ShanghaiTech Part A → MAE 237.69. Then tested pre-trained DM-Count models on an aerial rally photo to see how domain matters: 📸 rally.jpeg results: • DM-Count QNRF (high-res diverse): 2,077 I will soon publish this on GitHub. All running on MPS (no CUDA) 🔥 #ComputerVision #DeepLearning #CrowdCounting #PyTorch #MachineLearning