From c2de3a922bffecbcc06b30206d1ba2ce1c137b48 Mon Sep 17 00:00:00 2001 From: Ross Wightman Date: Sun, 26 May 2019 18:46:12 -0700 Subject: [PATCH] Update README.md --- README.md | 34 ++++++++++++++++++++++++++++++---- 1 file changed, 30 insertions(+), 4 deletions(-) diff --git a/README.md b/README.md index 4913a33e..bc676b8e 100644 --- a/README.md +++ b/README.md @@ -123,14 +123,40 @@ I've leveraged the training scripts in this repository to train a few of the mod | Single-Path NASNet 1.00 | 74.084 (25.916) | 91.818 (8.182) | 4.42M | bilinear | ### Ported Weights -|Model | Prec@1 (Err) | Prec@5 (Err) | Param # | Image Scaling | Source | + +#### @ 224x224 +| Model | Prec@1 (Err) | Prec@5 (Err) | Param # | Image Scaling | Source | +|---|---|---|---|---|---| +| senet154 | 81.224 (18.776) | 95.356 (4.644) | 115.09 | bicubic | | +| resnet152_v1s | 81.012 (18.988) | 95.416 (4.584) | 60.32 | bicubic | | +| seresnext101_32x4d | 80.902 (19.098) | 95.294 (4.706) | 48.96 | bicubic | | +| seresnext101_64x4d | 80.890 (19.110) | 95.304 (4.696) | 88.23 | bicubic | | +| resnext101_64x4d | 80.602 (19.398) | 94.994 (5.006) | 83.46 | bicubic | | +| resnet152_v1d | 80.470 (19.530) | 95.206 (4.794) | 60.21 | bicubic | | +| resnet101_v1d | 80.424 (19.576) | 95.020 (4.980) | 44.57 | bicubic | | +| resnext101_32x4d | 80.334 (19.666) | 94.926 (5.074) | 44.18 | bicubic | | +| resnet101_v1s | 80.300 (19.700) | 95.150 (4.850) | 44.67 | bicubic | | +| resnet152_v1c | 79.916 (20.084) | 94.842 (5.158) | 60.21 | bicubic | | +| seresnext50_32x4d | 79.912 (20.088) | 94.818 (5.182) | 27.56 | bicubic | | +| resnet152_v1b | 79.692 (20.308) | 94.738 (5.262) | 60.19 | bicubic | | +| resnet101_v1c | 79.544 (20.456) | 94.586 (5.414) | 44.57 | bicubic | | +| resnext50_32x4d | 79.356 (20.644) | 94.424 (5.576) | 25.03 | bicubic | | +| resnet101_v1b | 79.304 (20.696) | 94.524 (5.476) | 44.55 | bicubic | | +| resnet50_v1d | 79.074 (20.926) | 94.476 (5.524) | 25.58 | bicubic | | +| resnet50_v1s | 78.712 (21.288) | 94.242 (5.758) | 25.68 | bicubic | | +| resnet50_v1c | 78.010 (21.990) | 93.988 (6.012) | 25.58 | bicubic | | +| resnet50_v1b | 77.578 (22.422) | 93.718 (6.282) | 25.56 | bicubic | | +| resnet34_v1b | 74.580 (25.420) | 91.988 (8.012) | 21.80 | bicubic | | +| SE-MNASNet 1.00 (A1) | 73.086 (26.914) | 91.336 (8.664) | 3.87 | bicubic | [Google TFLite](https://github.com/tensorflow/tpu/tree/master/models/official/mnasnet) | +| MNASNet 1.00 (B1) | 72.398 (27.602) | 90.930 (9.070) | 4.36 | bicubic | [Google TFLite](https://github.com/tensorflow/tpu/tree/master/models/official/mnasnet) +| resnet18_v1b | 70.830 (29.170) | 89.756 (10.244) | 11.69 | bicubic | | + +#### @ 299x299 +| Model | Prec@1 (Err) | Prec@5 (Err) | Param # | Image Scaling | Source | |---|---|---|---|---|---| | Gluon Inception-V3 | 78.804 (21.196) | 94.380 (5.620) | 27.16M | bicubic | [MxNet Gluon](https://gluon-cv.mxnet.io/model_zoo/classification.html) | | Tensorflow Inception-V3 | 77.856 (22.144) | 93.644 (6.356) | 27.16M | bicubic | [Tensorflow Slim](https://github.com/tensorflow/models/tree/master/research/slim) | | Adversarially trained Inception-V3 | 77.576 (22.424) | 93.724 (6.276) | 27.16M | bicubic | [Tensorflow Adv models](https://github.com/tensorflow/models/tree/master/research/adv_imagenet_models) | -| SE-MNASNet 1.00 (A1) | 73.086 (26.914) | 91.336 (8.664) | 3.87M | bicubic | [Google TFLite](https://github.com/tensorflow/tpu/tree/master/models/official/mnasnet) | -| MNASNet 1.00 (B1) | 72.398 (27.602) | 90.930 (9.070) | 4.36M | bicubic | [Google TFLite](https://github.com/tensorflow/tpu/tree/master/models/official/mnasnet) | - NOTE: For some reason I can't hit the stated accuracy with my impl of MNASNet and Google's tflite weights. Using a TF equivalent to 'SAME' padding was important to get > 70%, but something small is still missing. Trying to train my own weights from scratch with these models has so far to leveled off in the same 72-73% range.