[FasterRCNN] using the wrong schedule for the published result

examples/FasterRCNN/README.md

@@ -47,7 +47,7 @@ To evaluate the performance (pretrained models can be downloaded in [model zoo](
 Mean Average Precision @IoU=0.50:0.95:
 
 + trainval35k/minival, FASTRCNN_BATCH=256: 34.2. Takes 49h on 8 TitanX.
-+ trainval35k/minival, FASTRCNN_BATCH=64: 32.7. Takes 25h on 8 TitanX.
++ trainval35k/minival, FASTRCNN_BATCH=64: 32.7. Takes 31h on 8 TitanX.
 
 The hyperparameters are not carefully tuned. You can probably get better performance by e.g. training longer.
 

examples/FasterRCNN/train.py

@@ -320,12 +320,12 @@ if __name__ == '__main__':
                     'learning_rate',
                     [(warmup_epoch * factor, 1e-2),
                      (150000 * factor // stepnum, 1e-3),
-                     (210000 * factor // stepnum, 1e-4)]),
+                     (230000 * factor // stepnum, 1e-4)]),
                 EvalCallback(),
                 GPUUtilizationTracker(),
             ],
             steps_per_epoch=stepnum,
-            max_epoch=230000 * factor // stepnum,
+            max_epoch=280000 * factor // stepnum,
             session_init=get_model_loader(args.load) if args.load else None,
         )
         trainer = SyncMultiGPUTrainerReplicated(get_nr_gpu())