[mingxiao huang, chainer] @nishino @kmaehashi hello, may you please help merge chainer/chainer#4933, chainer/chainer#5033 ? It has been pending for a long time. Thanks.
[mingxiao huang, chainer] @mitmul hello, how is your resnet50 training test going?
[Shunta Saito, chainer] Sorry I was on a business trip and too busy for another project preparing for CEATEC event next week, so couldn't have enough time to work on it...
[Shunta Saito, chainer] @mingxiao.huang Could you give us the evaluation script? I could find only the training scripts in this repository: https://github.com/mingxiaoh/chainer-training-script
[Shunta Saito, chainer] @mingxiao.huang Did you surely perform all of these techniques used in the inference time which are written in the original ResNet50 paper?
In testing, for comparison studies we adopt the standard 10-crop testing [21]. For best results, we adopt the fully convolutional form as in [41, 13], and average the scores at multiple scales (images are resized such that the shorter side is in {224, 256, 384, 480, 640}).
[mingxiao huang, chainer] @mitmul For your second question, as you can see the scripts we send to you, we used raw data images, then adopted the same data augmentation policy as caffe(https://github.com/intel/caffe/blob/master/models/intel_optimized_models/multinode/default_resnet50_8nodes/train_val.prototxt). With same HP, caffe can achieve SOTA accuracy. How about you send us directly the script you used for GPU training? Or would you please send the trained model to us, so that we can check where is the gap? Thanks.
[Shunta Saito, chainer] @mingxiao.huang I personally haven't run the full training of ResNet50 on ImageNet dataset so far. But I think someone has been done that with the official example code of Chainer found in https://github.com/chainer/chainer/tree/master/examples/imagenet . I'm trying to find the result, but haven't found that yet...
[Shunta Saito, chainer] @mingxiao.huang Well, could you share the log file of the training with Caffe that achieved "SOTA accuracy". Well, what do you mean by "SOTA"? I think ResNet50 is no longer the state-of-the-art in the image classification task.
[Shunta Saito, chainer] @mingxiao.huang Thanks. Well, how about the second question? What do you mean by “SOTA” with caffe… do you have any specific number in mind>
[mingxiao huang, chainer] IntelCaffe_16nodes_Tests_Train.log
[mingxiao huang, chainer] @mitmul I just sent you the log file, you can see that caffe acieved "top-1 = 0.75596, top-5 = 0.926322“
[Shunta Saito, chainer] @mingxiao.huang
From this part, Caffe does the inference time augmentation. You should do the same augmentation in a custom Evaluator class in the Chainer script.
https://github.com/intel/caffe/blob/master/models/intel_optimized_models/multinode/default_resnet50_8nodes/train_val.prototxt#L35-L63
From this part, Caffe does the inference time augmentation. You should do the same augmentation in a custom Evaluator class in the Chainer script.
https://github.com/intel/caffe/blob/master/models/intel_optimized_models/multinode/default_resnet50_8nodes/train_val.prototxt#L35-L63
[Shunta Saito, chainer] Doesn’t “random_resize_param” mean the random resizing augmentation?
[Shunta Saito, chainer] I think we cannot compare the results without unifying the preprocessing.
[mingxiao huang, chainer] @mitmul it means random resizing augmentation during training(not inference). Besides, according to our test(https://github.com/mingxiaoh/chainer-training-script/blob/master/train_imagenet_resnet50_blockmode_augmentation.py) on Chainer, it seems that there is no much difference whether we do augmentation or not.
[Shunta Saito, chainer] @mingxiao.huang OK thanks. I finally succeeded to allocate my time and the computational resource, so started the training script with 4 V100 GPUs. Please wait until it finishes. THanks
[Yong Wu, chainer] @nishino @mitmul Hi, We are working on a plan to support INT8/VNNI in Chainer/iDeep to accelerate inference on IA. We’d like to listen to your thoughts/plan on INT8 inference. Will it be part of Chainer 5.x roadmap? Or you think it should go with another path?
[mingxiao huang, chainer] @nishino wrt docker image, according to my understanding, latest-intel-python2 and latest-intel-python3 are generated by https://github.com/chainer/chainer/tree/master/docker/intel/python2 and https://github.com/chainer/chainer/tree/master/docker/intel/python3, respectively. But, where is the dockerfile for latest-intel? We downloaded the latest-intel and latest-intel-python2 images, and found that latest-intel is also based on Ubuntu 16.04.5 and 2.7.12 as latest-intel-python2. what is the difference between latest-intel and latest-intel-python2? Thanks.