The trained model provided by this repo only got mean_iou 0.7473?

[ksnzh]

(tf1.6-py35) ➜  tensorflow-deeplab-v3-plus git:(master) ✗ python evaluate.py --image_data_dir /media/kenzhang/DATA/VOC/VOCdevkit/VOC2012/JPEGImages --label_data_dir /media/kenzhang/DATA/VOC/VOC_aug/dataset/cls_png --model ./model
2018-03-26 20:49:57.031878: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:898] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2018-03-26 20:49:57.032225: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1212] Found device 0 with properties: 
name: GeForce GTX TITAN X major: 5 minor: 2 memoryClockRate(GHz): 1.076
pciBusID: 0000:01:00.0
totalMemory: 11.92GiB freeMemory: 10.62GiB
2018-03-26 20:49:57.032240: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1312] Adding visible gpu devices: 0
2018-03-26 20:49:57.206139: I tensorflow/core/common_runtime/gpu/gpu_device.cc:993] Creating TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 10284 MB memory) -> physical GPU (device: 0, name: GeForce GTX TITAN X, pci bus id: 0000:01:00.0, compute capability: 5.2)
INFO:tensorflow:Restoring parameters from ./model/model.ckpt-30358
INFO:tensorflow:current step = 100 (15.997 sec)
INFO:tensorflow:current step = 200 (11.710 sec)
INFO:tensorflow:current step = 300 (10.463 sec)
INFO:tensorflow:current step = 400 (10.616 sec)
INFO:tensorflow:current step = 500 (11.119 sec)
INFO:tensorflow:current step = 600 (10.345 sec)
INFO:tensorflow:current step = 700 (10.842 sec)
INFO:tensorflow:current step = 800 (11.057 sec)
INFO:tensorflow:current step = 900 (10.161 sec)
INFO:tensorflow:current step = 1000 (10.386 sec)
INFO:tensorflow:current step = 1100 (9.932 sec)
INFO:tensorflow:current step = 1200 (9.767 sec)
INFO:tensorflow:current step = 1300 (9.632 sec)
INFO:tensorflow:current step = 1400 (9.681 sec)
Intersection over Union for each class:
    class 0: 0.9227
    class 1: 0.8044
    class 2: 0.7312
    class 3: 0.8417
    class 4: 0.6443
    class 5: 0.7556
    class 6: 0.9166
    class 7: 0.7937
    class 8: 0.8811
    class 9: 0.4154
    class 10: 0.8246
    class 11: 0.5081
    class 12: 0.8194
    class 13: 0.8180
    class 14: 0.7963
    class 15: 0.7988
    class 16: 0.5321
    class 17: 0.8129
    class 18: 0.5304
    class 19: 0.8102
    class 20: 0.7368
mean Intersection over Union: 0.7473
Pixel Accuracy: 0.9340

[rishizek]

Hi @ksnzh , Thanks for your interest in the repo. Strange. I can get 0.7731 as follow:

$ python evaluate.py 
2018-03-26 22:43:39.285364: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:898] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2018-03-26 22:43:39.285874: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1212] Found device 0 with properties: 
name: GeForce GTX 1080 Ti major: 6 minor: 1 memoryClockRate(GHz): 1.582
pciBusID: 0000:01:00.0
totalMemory: 10.91GiB freeMemory: 10.56GiB
2018-03-26 22:43:39.285894: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1312] Adding visible gpu devices: 0
2018-03-26 22:43:39.859150: I tensorflow/core/common_runtime/gpu/gpu_device.cc:993] Creating TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 10219 MB memory) -> physical GPU (device: 0, name: GeForce GTX 1080 Ti, pci bus id: 0000:01:00.0, compute capability: 6.1)
INFO:tensorflow:Restoring parameters from ./model/model.ckpt-30358
INFO:tensorflow:current step = 100 (11.165 sec)
(snip...)
INFO:tensorflow:current step = 1400 (6.465 sec)
Intersection over Union for each class:
    class 0: 0.9408
    class 1: 0.8688
    class 2: 0.4256
    class 3: 0.8820
    class 4: 0.7174
    class 5: 0.8039
    class 6: 0.9385
    class 7: 0.8557
    class 8: 0.9159
    class 9: 0.4079
    class 10: 0.8778
    class 11: 0.5513
    class 12: 0.8621
    class 13: 0.8692
    class 14: 0.8401
    class 15: 0.8387
    class 16: 0.5838
    class 17: 0.8593
    class 18: 0.5640
    class 19: 0.8865
    class 20: 0.7447
mean Intersection over Union: 0.7731
Pixel Accuracy: 0.9478

Did you download the augmented segmentation data from here? That's the only potential difference that I can think of now.

[GeneralLi95]

@rishizek Excellent Job!