Randl / efficientdet-pytorch

A PyTorch impl of EfficientDet faithful to the original Google impl w/ ported weights

EfficientDet (PyTorch)

This is a work in progress PyTorch implementation of EfficientDet.

It is based on the

• official Tensorflow implementation by Mingxing Tan and the Google Brain team
• paper by Mingxing Tan, Ruoming Pang, Quoc V. Le EfficientDet: Scalable and Efficient Object Detection

I am aware there are other PyTorch implementations. Their approach didn't fit well with my aim to replicate the Tensorflow models closely enough to allow weight ports while still maintaining a PyTorch feel and a high degree of flexibility for future additions. So, this is built from scratch and leverages my previous EfficientNet work.

Updates / Tasks

2020-4-15

Taking a pause on training, some high priority things came up. There are signs of life on the training branch, was working the basic augs before priority switch, loss fn appeared to be doing something sane with distributed training working, no proper eval yet, init not correct yet. I will get to it, with SOTA training config and good performance as the end goal (as with my EfficientNet work).

2020-04-11

Cleanup post-processing. Less code and a five-fold throughput increase on the smaller models. D0 running > 130 img/s on a single 2080Ti, D1 > 130 img/s on dual 2080Ti up to D7 @ 8.5 img/s.

2020-04-10

Replace generate_detections with PyTorch impl using torchvision batched_nms. Significant performance increase with minor (+/-.001 mAP) score differences. Quite a bit faster than original TF impl on a GPU now.

2020-04-09

Initial code with working validation posted. Yes, it's a little slow, but I think faster than the official impl on a GPU if you leave AMP enabled. Post processing needs some love.

Core Tasks

• Feature extraction from my EfficientNet implementations (https://github.com/rwightman/gen-efficientnet-pytorch or https://github.com/rwightman/pytorch-image-models)
• Low level blocks / helpers (SeparableConv, create_pool2d (same padding), etc)
• PyTorch implementation of BiFPN, BoxNet, ClassNet modules and related submodules
• Port Tensorflow checkpoints to PyTorch -- initial D1 checkpoint converted, state_dict loaded, on to validation....
• Basic MS COCO validation script
  • Temporary (hacky) COCO dataset and transform
  • Port reference TF anchor and object detection code
  • Verify model output sanity
  • Integrate MSCOCO eval metric calcs
• Some cleanup, testing
• Submit to test-dev server, all good
• Add torch hub support and pretrained URL based weight download
• Change module dependencies from 'timm' to minimal 'geffnet' for backbone, bring some of the layers here leaving as timm for now, as the training code will use many timm functions that I leverage to reproduce SOTA EfficientNet training in PyTorch
• Remove redundant bias layers that exist in the official impl and weights
• Add visualization support
• Performance improvements, numpy TF detection code -> optimized PyTorch
• Verify/fix Torchscript and ONNX export compatibility

Possible Future Tasks

• Training (object detection) reimplementation w/ Rand/AutoAugment, etc
• Training (semantic segmentation) experiments
• Integration with Detectron2 / MMDetection codebases
• Addition and cleanup of EfficientNet based U-Net and DeepLab segmentation models that I've used in past projects
• Addition and cleanup of OpenImages dataset/training support from a past project
• Exploration of instance segmentation possibilities...

If you are an organization is interested in sponsoring and any of this work, or prioritization of the possible future directions interests you, feel free to contact me (issue, LinkedIn, Twitter, hello at rwightman dot com). I will setup a github sponser if there is any interest.

Models

Variant	Download	mAP (val2017)	mAP (test-dev2017)	mAP (Tensorflow official test-dev2017)
D0	tf_efficientdet_d0.pth	32.8	TBD	33.8
D1	tf_efficientdet_d1.pth	38.5	TBD	39.6
D2	tf_efficientdet_d2.pth	42.0	42.5	43
D3	tf_efficientdet_d3.pth	45.3	TBD	45.8
D4	tf_efficientdet_d4.pth	48.3	TBD	49.4
D5	tf_efficientdet_d5.pth	49.6	TBD	50.7
D6	tf_efficientdet_d6.pth	50.6	TBD	51.7
D7	tf_efficientdet_d7.pth	50.9	51.2	52.2

Usage

Environment Setup

Tested in a Python 3.7 or 3.8 conda environment in Linux with:

• PyTorch 1.4
• PyTorch Image Models (timm) 0.1.20, pip install timm or local install from (https://github.com/rwightman/pytorch-image-models)
• Apex AMP master (as of 2020-04)

NOTE - There is a conflict/bug with Numpy 1.18+ and pycocotools, force install numpy <= 1.17.5 or the coco eval will fail, the validation script will still save the output JSON and that can be run through eval again later.

Dataset Setup

MSCOCO 2017 validation data:

wget http://images.cocodataset.org/zips/val2017.zip
wget http://images.cocodataset.org/annotations/annotations_trainval2017.zip
unzip val2017.zip
unzip annotations_trainval2017.zip

MSCOCO 2017 test-dev data:

wget http://images.cocodataset.org/zips/test2017.zip
unzip -q test2017.zip
wget http://images.cocodataset.org/annotations/image_info_test2017.zip
unzip image_info_test2017.zip

Run COCO Evaluation

Run validation (val2017 by default) with D2 model: python validation.py /localtion/of/mscoco/ --model tf_efficientdet_d2 --checkpoint tf_efficientdet_d2.pth

Run test-dev2017: python validation.py /localtion/of/mscoco/ --model tf_efficientdet_d2 --checkpoint tf_efficientdet_d2.pth --anno test-dev2017

Run Inference

TODO: Need an inference script

Results

TEST-DEV2017

NOTE: I've only tried submitting D2 and D7 to dev server for sanity check so far

EfficientDet-D2

 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.425
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.618
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.457
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.222
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.467
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.578
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.341
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.536
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.569
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.334
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.627
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.749

EfficientDet-D7

 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.512
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.707
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.554
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.335
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.549
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.639
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.385
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.623
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.660
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.485
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.696
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.800

VAL2017

EfficientDet-D0

 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.328
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.509
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.346
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.118
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.380
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.520
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.283
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.432
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.457
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.185
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.539
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.679

EfficientDet-D1

 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.385
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.575
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.410
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.179
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.441
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.561
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.318
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.491
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.519
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.278
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.589
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.721

EfficientDet-D2

 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.420
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.612
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.448
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.232
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.476
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.582
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.338
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.531
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.562
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.340
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.627
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.741

EfficientDet-D3

 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.453
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.646
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.485
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.272
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.498
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.614
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.355
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.563
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.597
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.399
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.647
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.762

EfficientDet-D4

Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.483
Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.680
Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.521
Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.325
Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.535
Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.634
Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.371
Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.591
Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.626
Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.453
Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.676
Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.772

EfficientDet-D5

 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.496
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.692
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.533
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.327
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.543
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.642
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.377
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.608
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.644
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.469
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.688
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.785

EfficientDet-D6

 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.506
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.701
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.545
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.336
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.552
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.650
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.382
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.616
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.653
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.477
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.697
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.791

EfficientDet-D7

Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.509
Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.705
Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.548
Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.353
Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.559
Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.648
Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.381
Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.617
Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.654
Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.489
Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.696
Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.791