# Mean image for imagenet model (e.g., vgg16)

**URL:** <https://discuss.pytorch.org/t/mean-image-for-imagenet-model-e-g-vgg16/3787>\
**Category:** Uncategorized\
**Created:** [June 6, 2017, 11:55pm UTC](https://discuss.pytorch.org/t/mean-image-for-imagenet-model-e-g-vgg16/3787 "2017-06-06T23:55:13Z")\
**Posts on this page:** 4\
**Page:** 1

<div class="post-metadata">

**Author:** ![johnny5550822](https://discuss.pytorch.org/user_avatar/discuss.pytorch.org/johnny5550822/32/497_2.png) [@johnny5550822](https://discuss.pytorch.org/u/johnny5550822)\
**Post date:** [June 6, 2017, 11:55pm UTC](https://discuss.pytorch.org/t/mean-image-for-imagenet-model-e-g-vgg16/3787/1 "2017-06-06T23:55:14Z")

</div>

Are there any mean image file for the imagenet model such as vgg16?

---

<div class="post-metadata">

**Author:** ![panovr](https://discuss.pytorch.org/user_avatar/discuss.pytorch.org/panovr/32/624_2.png) [@panovr](https://discuss.pytorch.org/u/panovr)\
**Post date:** [June 7, 2017, 1:37am UTC](https://discuss.pytorch.org/t/mean-image-for-imagenet-model-e-g-vgg16/3787/2 "2017-06-07T01:37:36Z")

</div>

From PyTorch ImageNet training script, you can find the mean value at [here](https://github.com/pytorch/examples/blob/master/imagenet/main.py#L100).

---

<div class="post-metadata">

**Author:** ![johnny5550822](https://discuss.pytorch.org/user_avatar/discuss.pytorch.org/johnny5550822/32/497_2.png) [@johnny5550822](https://discuss.pytorch.org/u/johnny5550822)\
**Post date:** [June 7, 2017, 1:38am UTC](https://discuss.pytorch.org/t/mean-image-for-imagenet-model-e-g-vgg16/3787/3 "2017-06-07T01:38:58Z")

</div>

Thanks. But I think this is just the overall mean for normalization, but not the mean image? (3x256x256)

---

<div class="post-metadata">

**Author:** ![smth](https://discuss.pytorch.org/user_avatar/discuss.pytorch.org/smth/32/13_2.png) [@smth](https://discuss.pytorch.org/u/smth)\
**Post date:** [June 21, 2017, 10:10pm UTC](https://discuss.pytorch.org/t/mean-image-for-imagenet-model-e-g-vgg16/3787/4 "2017-06-21T22:10:03Z")

</div>

we dont have a full mean image (as it is not necessary to achieve convergence). We just compute one global mean value per image channel, instead of a mean image.