Fix train.py for saving sample npy (#201)

scripts/evaluate_svhn_tinierssd.sh

@@ -0,0 +1,2 @@
+#!/bin/sh
+python train.py --deterministic --print-freq 200 --model ai85tinierssd --use-bias --dataset SVHN_74 --device MAX78000 --obj-detection --obj-detection-params parameters/obj_detection_params_svhn.yaml --qat-policy policies/qat_policy_svhn.yaml --evaluate -8 --exp-load-weights-from ../ai8x-synthesis/trained/ai85-svhn-tinierssd-qat8-q.pth.tar "$@"

train.py

@@ -194,6 +194,10 @@ def main():
             print('WARNING: Initial learning rate (--lr) not set, selecting 0.1.')
         args.lr = 0.1
 
+    if args.generate_sample is not None and not args.act_mode_8bit:
+        print('WARNING: Cannot save sample in training mode, ignoring --save-sample option. '
+              'Use with --evaluate instead.')
+
     msglogger = apputils.config_pylogger(os.path.join(script_dir, 'logging.conf'), args.name,
                                          args.output_dir)
 
@@ -1020,6 +1024,7 @@ def save_tensor(t, f, regression=True):
     end = time.time()
     class_probs = []
     class_preds = []
+    sample_saved = False  # Track if --save-sample has been done for this validation step
 
     # Get object detection params
     obj_detection_params = parse_obj_detection_yaml.parse(args.obj_detection_params) \
@@ -1091,9 +1096,9 @@ def save_tensor(t, f, regression=True):
                                 and model.__dict__['_modules'][key].wide):
                             output /= 256.
 
-            if args.generate_sample is not None:
+            if args.generate_sample is not None and args.act_mode_8bit and not sample_saved:
                 sample.generate(args.generate_sample, inputs, target, output, args.dataset, False)
-                return .0, .0, .0, .0
+                sample_saved = True
 
             if args.csv_prefix is not None:
                 save_tensor(inputs, f_x)