Add plots of validation error, training error, and training loss

See #18
facebookarchive · Mar 4, 2016 · 0663aba · 0663aba
1 parent eff127f
commit 0663aba
Show file tree

Hide file tree

Showing 2 changed files with 17 additions and 0 deletions.
diff --git a/pretrained/CONVERGENCE.md b/pretrained/CONVERGENCE.md
@@ -0,0 +1,14 @@
+These are plots of validation error, training error, and training loss on ImageNet measured after every epoch. 
+
+- [Validation Error](#validation-error)
+- [Training Error](#training-error)
+- [Training Loss](#training-loss)
+
+## Validation Error
+<img width="842" alt="validation error" src="https://cloud.githubusercontent.com/assets/655866/13538067/6785a948-e216-11e5-8f76-b74e74397fd7.png">
+
+## Training Error
+<img width="828" alt="train error" src="https://cloud.githubusercontent.com/assets/655866/13538065/6775ec38-e216-11e5-831d-26e5660367a5.png">
+
+## Training Loss
+<img width="836" alt="train loss" src="https://cloud.githubusercontent.com/assets/655866/13538066/678459b2-e216-11e5-8b13-88c24d73ab5d.png">
diff --git a/pretrained/README.md b/pretrained/README.md
@@ -28,6 +28,9 @@ The ResNet-50 model has a batch normalization layer after the addition, instead
 | ResNet-50     | 22.24       | 6.08        |
 | ResNet-101    | **21.08**   | **5.35**    |
 
+##### ImageNet charts
+
+See the [convergence plots](CONVERGENCE.md) for charts of training and validation error and training loss after every epoch.
 
 ### Fine-tuning on a custom dataset