Improved Distribution Matching for Dataset Condensation