Add pretrained weights for non tf efficientdet_d1 and tf_efficientdet_lite0

Author: rwightman

README.md

@@ -18,6 +18,33 @@ Aside from the default model configs, there is a lot of flexibility to facilitat
 
 ## Updates / Tasks
 
+### 2020-06-14
+New model results, I've trained a D1 model with some WIP augmentation enhancements (not commited), just squeaking by official weights.
+
+EfficientDet-D1:
+```
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.393798
+ Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.586831
+ Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.420305
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.191880
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.455586
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.571316
+```
+
+Also, [Soyeb Nagori](https://github.com/soyebn) trained an EffiCientDet-Lite0 config using this code and contributed the weights.
+```
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.319861
+ Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.500062
+ Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.336777
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.111257
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.378062
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.501938
+```
+
+Unlike the other tf_ prefixed models this is not ported from (as of yet unreleased) TF official model, but it used
+TF ported weights for the pretrained imagenet model that was the starting point, thus it uses SAME padding. 
+
+
 ### 2020-06-12
 
 * Additional experimental model configs based on MobileNetV2, MobileNetV3, MixNet, EfficientNet-Lite. Requires
@@ -40,12 +67,6 @@ My latest D0 run:
  Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.123988
  Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.395033
  Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.521695
- Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.287121
- Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.441450
- Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.467914
- Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.197697
- Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.552515
- Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.689297
 ```
 
 TF ported D0 weights:
@@ -56,12 +77,6 @@ TF ported D0 weights:
  Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.125278
  Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.386957
  Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.528071
- Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.288049
- Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.439918
- Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.466877
- Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.193482
- Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.549262
- Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.686037
 ```
 
 Pretrained weights added for this model `efficientdet_d0`  (Tensorflow port is `tf_efficientdet_d0`)
@@ -138,9 +153,11 @@ If you are an organization is interested in sponsoring and any of this work, or
 
 | Variant | Download | mAP (val2017) | mAP (test-dev2017) | mAP (TF official val2017) | mAP (TF official test-dev2017) |
 | --- | --- | :---: | :---: | :---: | :---: |
+| lite0 | [tf_efficientdet_lite0.pth](https://github.com/rwightman/efficientdet-pytorch/releases/download/v0.1/tf_efficientdet_lite0-f5f303a9.pth) | 32.0 | TBD | N/A | N/A |
 | D0 | [efficientdet_d0.pth](https://github.com/rwightman/efficientdet-pytorch/releases/download/v0.1/efficientdet_d0-f3276ba8.pth) | 33.6 | TBD | 33.5 | 33.8 |
 | D0 | [tf_efficientdet_d0.pth](https://github.com/rwightman/efficientdet-pytorch/releases/download/v0.1/tf_efficientdet_d0-d92fd44f.pth) | 33.6 | TBD | 33.5 | 33.8 |
 | D1 | [tf_efficientdet_d1.pth](https://github.com/rwightman/efficientdet-pytorch/releases/download/v0.1/tf_efficientdet_d1-4c7ebaf2.pth) | 39.3 | TBD | 39.1 | 39.6 |
+| D1 | [efficientdet_d1.pth](https://github.com/rwightman/efficientdet-pytorch/releases/download/v0.1/efficientdet_d1-bb7e98fe.pth) | 39.4 | TBD | 39.1 | 39.6 |
 | D2 | [tf_efficientdet_d2.pth](https://github.com/rwightman/efficientdet-pytorch/releases/download/v0.1/tf_efficientdet_d2-cb4ce77d.pth) | 42.6 | 43.1 | 42.5 | 43 |
 | D3 | [tf_efficientdet_d3.pth](https://github.com/rwightman/efficientdet-pytorch/releases/download/v0.1/tf_efficientdet_d3-b0ea2cbc.pth) | 46.0 | TBD | 45.9 | 45.8 |
 | D4 | [tf_efficientdet_d4.pth](https://github.com/rwightman/efficientdet-pytorch/releases/download/v0.1/tf_efficientdet_d4-5b370b7a.pth) | 49.1 | TBD | 49.0 | 49.4 |

effdet/config/model_config.py

@@ -85,7 +85,7 @@ def default_detection_model_configs():
         pad_type='',
         redundant_bias=False,
         backbone_args=dict(drop_path_rate=0.2),
-        url='',  # no pretrained weights yet
+        url='https://github.com/rwightman/efficientdet-pytorch/releases/download/v0.1/efficientdet_d1-bb7e98fe.pth',
     ),
     efficientdet_d2=dict(
         name='efficientdet_d2',
@@ -286,8 +286,11 @@ def default_detection_model_configs():
         fpn_cell_repeats=3,
         box_class_repeats=3,
         act_type='relu',
+        redundant_bias=False,
         backbone_args=dict(drop_path_rate=0.1),
-        url='',  # no pretrained weights yet
+        # unlike other tf_ models, this was not ported from tf automl impl, but trained from tf pretrained efficient lite
+        # weights using this code, will likely replace if/when official det-lite weights are released
+        url='https://github.com/rwightman/efficientdet-pytorch/releases/download/v0.1/tf_efficientdet_lite0-f5f303a9.pth',
     ),
     tf_efficientdet_lite1=dict(
         name='tf_efficientdet_lite1',