Merge branch 'enhanced_validate.py' of github.com:ha405/pytorch-image-models into ha405-enhanced_validate.py

Author: rwightman

validate.py

@@ -42,6 +42,12 @@
 
 has_compile = hasattr(torch, 'compile')
 
+try:
+    from sklearn.metrics import precision_score, recall_score, f1_score
+    has_sklearn = True
+except ImportError:
+    has_sklearn = False
+
 _logger = logging.getLogger('validate')
 
 
@@ -158,6 +164,11 @@
 parser.add_argument('--retry', default=False, action='store_true',
                     help='Enable batch size decay & retry for single model validation')
 
+parser.add_argument('--metrics-avg', type=str, default=None,
+                    choices=['micro', 'macro', 'weighted'],
+                    help='Enable precision, recall, F1-score calculation and specify the averaging method. '
+                         'Requires scikit-learn. (default: None)')
+
 # NaFlex loader arguments
 parser.add_argument('--naflex-loader', action='store_true', default=False,
                    help='Use NaFlex loader (Requires NaFlex compatible model)')
@@ -176,6 +187,11 @@ def validate(args):
 
     device = torch.device(args.device)
 
+    if args.metrics_avg and not has_sklearn:
+        _logger.warning(
+            f"scikit-learn not installed, disabling metrics calculation. Please install with 'pip install scikit-learn'.")
+        args.metrics_avg = None
+
     model_dtype = None
     if args.model_dtype:
         assert args.model_dtype in ('float32', 'float16', 'bfloat16')
@@ -346,6 +362,10 @@ def validate(args):
     top1 = AverageMeter()
     top5 = AverageMeter()
 
+    if args.metrics_avg:
+        all_preds = []
+        all_targets = []
+
     model.eval()
     with torch.inference_mode():
         # warmup, reduce variability of first batch time, especially for comparing torchscript vs non
@@ -382,6 +402,11 @@ def validate(args):
             top1.update(acc1.item(), batch_size)
             top5.update(acc5.item(), batch_size)
 
+            if args.metrics_avg:
+                predictions = torch.argmax(output, dim=1)
+                all_preds.append(predictions.cpu())
+                all_targets.append(target.cpu())
+
             # measure elapsed time
             batch_time.update(time.time() - end)
             end = time.time()
@@ -408,18 +433,41 @@ def validate(args):
         top1a, top5a = real_labels.get_accuracy(k=1), real_labels.get_accuracy(k=5)
     else:
         top1a, top5a = top1.avg, top5.avg
+
+    metric_results = {}
+    if args.metrics_avg:
+        all_preds = torch.cat(all_preds).numpy()
+        all_targets = torch.cat(all_targets).numpy()
+        precision = precision_score(all_targets, all_preds, average=args.metrics_avg, zero_division=0)
+        recall = recall_score(all_targets, all_preds, average=args.metrics_avg, zero_division=0)
+        f1 = f1_score(all_targets, all_preds, average=args.metrics_avg, zero_division=0)
+        metric_results = {
+            f'{args.metrics_avg}_precision': round(precision, 4),
+            f'{args.metrics_avg}_recall': round(recall, 4),
+            f'{args.metrics_avg}_f1_score': round(f1, 4),
+        }
+
     results = OrderedDict(
         model=args.model,
         top1=round(top1a, 4), top1_err=round(100 - top1a, 4),
         top5=round(top5a, 4), top5_err=round(100 - top5a, 4),
+        **metric_results,
         param_count=round(param_count / 1e6, 2),
         img_size=data_config['input_size'][-1],
         crop_pct=crop_pct,
         interpolation=data_config['interpolation'],
     )
 
-    _logger.info(' * Acc@1 {:.3f} ({:.3f}) Acc@5 {:.3f} ({:.3f})'.format(
-       results['top1'], results['top1_err'], results['top5'], results['top5_err']))
+    log_string = ' * Acc@1 {:.3f} ({:.3f}) Acc@5 {:.3f} ({:.3f})'.format(
+       results['top1'], results['top1_err'], results['top5'], results['top5_err'])
+    if metric_results:
+        log_string += ' | Precision({avg}) {prec:.3f} | Recall({avg}) {rec:.3f} | F1-score({avg}) {f1:.3f}'.format(
+            avg=args.metrics_avg,
+            prec=metric_results[f'{args.metrics_avg}_precision'],
+            rec=metric_results[f'{args.metrics_avg}_recall'],
+            f1=metric_results[f'{args.metrics_avg}_f1_score'],
+        )
+    _logger.info(log_string)
 
     return results
 
@@ -534,4 +582,4 @@ def write_results(results_file, results, format='csv'):
 
 
 if __name__ == '__main__':
-    main()
+    main()