Skip to content

Commit 8fc03d4

Browse files
rwightmanrwightman
committed
Port and updated D7 weights from official TF repo (53.1 mAP in PyTorch). Fix a train arg desc.
1 parent 4c4be27 commit 8fc03d4

File tree

3 files changed

+32
-29
lines changed

3 files changed

+32
-29
lines changed

‎README.md

Lines changed: 30 additions & 27 deletions
Original file line numberDiff line numberDiff line change
@@ -18,6 +18,9 @@ Aside from the default model configs, there is a lot of flexibility to facilitat
1818

1919
## Updates / Tasks
2020

21+
### 2020-06-15
22+
Add updated D7 weights from Tensorflow impl, 53.1 validation mAP here (53.4 in TF)
23+
2124
### 2020-06-14
2225
New model results, I've trained a D1 model with some WIP augmentation enhancements (not commited), just squeaking by official weights.
2326

@@ -154,16 +157,16 @@ If you are an organization is interested in sponsoring and any of this work, or
154157
| Variant | Download | mAP (val2017) | mAP (test-dev2017) | mAP (TF official val2017) | mAP (TF official test-dev2017) |
155158
| --- | --- | :---: | :---: | :---: | :---: |
156159
| 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 |
157-
| 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 |
158160
| 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 |
161+
| 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 |
159162
| 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 |
160-
| 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 |
163+
| D1 | [efficientdet_d1.pth](https://github.com/rwightman/efficientdet-pytorch/releases/download/v0.1/efficientdet_d1-bb7e98fe.pth) | 39.4 | 39.5 | 39.1 | 39.6 |
161164
| 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 |
162165
| 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 |
163166
| 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 |
164167
| D5 | [tf_efficientdet_d5.pth](https://github.com/rwightman/efficientdet-pytorch/releases/download/v0.1/tf_efficientdet_d5-ef44aea8.pth) | 50.4 | TBD | 50.5 | 50.7 |
165168
| D6 | [tf_efficientdet_d6.pth](https://github.com/rwightman/efficientdet-pytorch/releases/download/v0.1/tf_efficientdet_d6-51cb0132.pth) | 51.2 | TBD | 51.3 | 51.7 |
166-
| D7 | [tf_efficientdet_d7.pth](https://github.com/rwightman/efficientdet-pytorch/releases/download/v0.1/tf_efficientdet_d7-f05bf714.pth) | 51.8 | 52.1 | 52.1 | 52.2 |
169+
| D7 | [tf_efficientdet_d7.pth](https://github.com/rwightman/efficientdet-pytorch/releases/download/v0.1/tf_efficientdet_d7_53-6d1d7a95.pth) | 53.1 | 53.4 | 53.4 | 53.7 |
167170

168171

169172
## Usage
@@ -287,18 +290,18 @@ NOTE: I've only tried submitting D2 and D7 to dev server for sanity check so far
287290

288291
##### EfficientDet-D7
289292
```
290-
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.521
291-
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.714
292-
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.563
293-
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.345
294-
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.555
295-
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.646
296-
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.390
297-
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.631
298-
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.670
299-
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.497
300-
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.704
301-
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.808
293+
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.534
294+
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.726
295+
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.577
296+
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.356
297+
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.569
298+
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.660
299+
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.397
300+
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.644
301+
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.682
302+
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.508
303+
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.718
304+
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.818
302305
```
303306

304307
#### VAL2017
@@ -417,16 +420,16 @@ NOTE: I've only tried submitting D2 and D7 to dev server for sanity check so far
417420

418421
##### EfficientDet-D7
419422
```
420-
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.518
421-
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.711
422-
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.558
423-
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.368
424-
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.564
425-
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.655
426-
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.386
427-
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.627
428-
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.665
429-
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.505
430-
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.704
431-
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.801
423+
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.531256
424+
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.724700
425+
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.571787
426+
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.368872
427+
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.573938
428+
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.668253
429+
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.393620
430+
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.637601
431+
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.676987
432+
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.524850
433+
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.717553
434+
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.806352
432435
```

‎effdet/config/model_config.py

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -274,7 +274,7 @@ def default_detection_model_configs():
274274
anchor_scale=5.0,
275275
fpn_name='bifpn_sum', # Use unweighted sum for training stability.
276276
backbone_args=dict(drop_path_rate=0.2),
277-
url='https://github.com/rwightman/efficientdet-pytorch/releases/download/v0.1/tf_efficientdet_d7-f05bf714.pth'
277+
url='https://github.com/rwightman/efficientdet-pytorch/releases/download/v0.1/tf_efficientdet_d7_53-6d1d7a95.pth'
278278
),
279279

280280
# The lite configs are in TF automl repository but no weights yet and listed as 'not final'

‎train.py

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -97,7 +97,7 @@ def add_bool_arg(parser, name, default=False, help=''): # FIXME move to utils
9797
parser.add_argument('--opt', default='momentum', type=str, metavar='OPTIMIZER',
9898
help='Optimizer (default: "momentum"')
9999
parser.add_argument('--opt-eps', default=1e-3, type=float, metavar='EPSILON',
100-
help='Optimizer Epsilon (default: 1e-8)')
100+
help='Optimizer Epsilon (default: 1e-3)')
101101
parser.add_argument('--momentum', type=float, default=0.9, metavar='M',
102102
help='SGD momentum (default: 0.9)')
103103
parser.add_argument('--weight-decay', type=float, default=4e-5,

0 commit comments

Comments
 (0)