pengzhao-intel commented on issue #12808: MKL-DNN Quantization Examples and README URL: https://github.com/apache/incubator-mxnet/pull/12808#issuecomment-430537478 @aaronmarkham I asked another member in our team to reproduce all results only following the README. And all results are reproducible smoothly :) The log as below, a little long. Could you help merge this PR? Thanks :) > ResNet50-V1 > > INFO:logger:batch size = 64 for inference > INFO:logger:rgb_mean = 123.68,116.779,103.939 > INFO:logger:rgb_std = 58.393,57.12,57.375 > INFO:logger:label_name = softmax_label > INFO:logger:Input data shape = (3, 224, 224) > INFO:logger:Dataset for inference: ./data/val_256_q90.rec > [11:00:02] src/io/iter_image_recordio_2.cc:172: ImageRecordIOParser2: ./data/val_256_q90.rec, use 1 threads for decoding.. > INFO:logger:Loading symbol from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/resnet50_v1-symbol.json > INFO:logger:Loading params from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/resnet50_v1-0000.params > INFO:logger:Skipping the first 50 batches > INFO:logger:Running model ./model/resnet50_v1-symbol.json for inference > [11:00:07] src/operator/subgraph/mkldnn/mkldnn_conv_property.cc:138: Start to execute MKLDNN Convolution optimization pass. > [11:00:07] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 147456 bytes with malloc directly > [11:00:07] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 589824 bytes with malloc directly > [11:00:08] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 2359296 bytes with malloc directly > [11:00:08] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 9437184 bytes with malloc directly > INFO:logger:Finished inference with 32000 images > INFO:logger:Finished with 95.636832 images per second > INFO:logger:('accuracy', 0.76490625) > INFO:logger:('top_k_accuracy_5', 0.93184375) > > INFO:logger:batch size = 64 for inference > INFO:logger:rgb_mean = 123.68,116.779,103.939 > INFO:logger:rgb_std = 58.393,57.12,57.375 > INFO:logger:label_name = softmax_label > INFO:logger:Input data shape = (3, 224, 224) > INFO:logger:Dataset for inference: ./data/val_256_q90.rec > [11:06:44] src/io/iter_image_recordio_2.cc:172: ImageRecordIOParser2: ./data/val_256_q90.rec, use 1 threads for decoding.. > INFO:logger:Loading symbol from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/resnet50_v1-quantized-5batches-naive-symbol.json > INFO:logger:Loading params from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/resnet50_v1-quantized-0000.params > INFO:logger:Skipping the first 50 batches > INFO:logger:Running model ./model/resnet50_v1-quantized-5batches-naive-symbol.json for inference > [11:06:49] src/operator/subgraph/mkldnn/mkldnn_conv_property.cc:138: Start to execute MKLDNN Convolution optimization pass. > [11:06:50] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 147456 bytes with malloc directly > [11:06:50] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 589824 bytes with malloc directly > [11:06:51] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 2359296 bytes with malloc directly > INFO:logger:Finished inference with 32000 images > INFO:logger:Finished with 47.259650 images per second > INFO:logger:('accuracy', 0.76096875) > INFO:logger:('top_k_accuracy_5', 0.92978125) > > INFO:logger:batch size = 64 for inference > INFO:logger:rgb_mean = 0,0,0 > INFO:logger:rgb_std = 1,1,1 > INFO:logger:label_name = softmax_label > INFO:logger:Input data shape = (3, 224, 224) > INFO:logger:Running model ./model/resnet50_v1-symbol.json for inference > INFO:logger:Loading symbol from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/resnet50_v1-symbol.json > /home/junmingc/xinyu-pr/incubator-mxnet/python/mxnet/module/base_module.py:68: UserWarning: Data provided by label_shapes don't match names specified by label_names ([] vs. ['softmax_label']) > warnings.warn(msg) > [12:50:25] src/operator/subgraph/mkldnn/mkldnn_conv_property.cc:138: Start to execute MKLDNN Convolution optimization pass. > [12:50:25] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 147456 bytes with malloc directly > [12:50:25] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 589824 bytes with malloc directly > [12:50:25] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 2359296 bytes with malloc directly > [12:50:25] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 9437184 bytes with malloc directly > INFO:logger:batch size 64, image/sec: 413.033026 > INFO:logger:batch size = 64 for inference > INFO:logger:rgb_mean = 0,0,0 > INFO:logger:rgb_std = 1,1,1 > INFO:logger:label_name = softmax_label > INFO:logger:Input data shape = (3, 224, 224) > INFO:logger:Running model ./model/resnet50_v1-quantized-5batches-naive-symbol.json for inference > INFO:logger:Loading symbol from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/resnet50_v1-quantized-5batches-naive-symbol.json > /home/junmingc/xinyu-pr/incubator-mxnet/python/mxnet/module/base_module.py:68: UserWarning: Data provided by label_shapes don't match names specified by label_names ([] vs. ['softmax_label']) > warnings.warn(msg) > [12:51:46] src/operator/subgraph/mkldnn/mkldnn_conv_property.cc:138: Start to execute MKLDNN Convolution optimization pass. > [12:51:47] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 147456 bytes with malloc directly > [12:51:47] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 589824 bytes with malloc directly > [12:51:47] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 2359296 bytes with malloc directly > INFO:logger:batch size 64, image/sec: 653.498226 > > ResNet101-V1 > > > INFO:logger:batch size = 64 for inference > INFO:logger:rgb_mean = 123.68,116.779,103.939 > INFO:logger:rgb_std = 58.393,57.12,57.375 > INFO:logger:label_name = softmax_label > INFO:logger:Input data shape = (3, 224, 224) > INFO:logger:Dataset for inference: ./data/val_256_q90.rec > [13:19:48] src/io/iter_image_recordio_2.cc:172: ImageRecordIOParser2: ./data/val_256_q90.rec, use 1 threads for decoding.. > INFO:logger:Loading symbol from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/resnet101_v1-symbol.json > INFO:logger:Loading params from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/resnet101_v1-0000.params > INFO:logger:Skipping the first 50 batches > INFO:logger:Running model ./model/resnet101_v1-symbol.json for inference > [13:19:53] src/operator/subgraph/mkldnn/mkldnn_conv_property.cc:138: Start to execute MKLDNN Convolution optimization pass. > [13:19:53] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 147456 bytes with malloc directly > [13:19:53] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 589824 bytes with malloc directly > [13:19:53] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 2359296 bytes with malloc directly > [13:19:54] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 9437184 bytes with malloc directly > INFO:logger:Finished inference with 32000 images > INFO:logger:Finished with 188.447936 images per second > INFO:logger:('accuracy', 0.773) > INFO:logger:('top_k_accuracy_5', 0.9358125) > INFO:logger:batch size = 64 for inference > INFO:logger:rgb_mean = 123.68,116.779,103.939 > INFO:logger:rgb_std = 58.393,57.12,57.375 > INFO:logger:label_name = softmax_label > INFO:logger:Input data shape = (3, 224, 224) > INFO:logger:Dataset for inference: ./data/val_256_q90.rec > [13:22:45] src/io/iter_image_recordio_2.cc:172: ImageRecordIOParser2: ./data/val_256_q90.rec, use 1 threads for decoding.. > INFO:logger:Loading symbol from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/resnet101_v1-quantized-5batches-naive-symbol.json > INFO:logger:Loading params from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/resnet101_v1-quantized-0000.params > INFO:logger:Skipping the first 50 batches > INFO:logger:Running model ./model/resnet101_v1-quantized-5batches-naive-symbol.json for inference > [13:22:50] src/operator/subgraph/mkldnn/mkldnn_conv_property.cc:138: Start to execute MKLDNN Convolution optimization pass. > [13:22:50] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 147456 bytes with malloc directly > [13:22:50] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 589824 bytes with malloc directly > [13:22:51] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 2359296 bytes with malloc directly > INFO:logger:Finished inference with 32000 images > INFO:logger:Finished with 265.491600 images per second > INFO:logger:('accuracy', 0.77096875) > INFO:logger:('top_k_accuracy_5', 0.93415625) > INFO:logger:batch size = 64 for inference > INFO:logger:rgb_mean = 0,0,0 > INFO:logger:rgb_std = 1,1,1 > INFO:logger:label_name = softmax_label > INFO:logger:Input data shape = (3, 224, 224) > INFO:logger:Running model ./model/resnet101_v1-symbol.json for inference > INFO:logger:Loading symbol from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/resnet101_v1-symbol.json > /home/junmingc/xinyu-pr/incubator-mxnet/python/mxnet/module/base_module.py:68: UserWarning: Data provided by label_shapes don't match names specified by label_names ([] vs. ['softmax_label']) > warnings.warn(msg) > [13:24:53] src/operator/subgraph/mkldnn/mkldnn_conv_property.cc:138: Start to execute MKLDNN Convolution optimization pass. > [13:24:53] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 147456 bytes with malloc directly > [13:24:53] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 589824 bytes with malloc directly > [13:24:53] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 2359296 bytes with malloc directly > [13:24:54] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 9437184 bytes with malloc directly > INFO:logger:batch size 64, image/sec: 235.930348 > INFO:logger:batch size = 64 for inference > INFO:logger:rgb_mean = 0,0,0 > INFO:logger:rgb_std = 1,1,1 > INFO:logger:label_name = softmax_label > INFO:logger:Input data shape = (3, 224, 224) > INFO:logger:Running model ./model/resnet101_v1-quantized-5batches-naive-symbol.json for inference > INFO:logger:Loading symbol from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/resnet101_v1-quantized-5batches-naive-symbol.json > /home/junmingc/xinyu-pr/incubator-mxnet/python/mxnet/module/base_module.py:68: UserWarning: Data provided by label_shapes don't match names specified by label_names ([] vs. ['softmax_label']) > warnings.warn(msg) > [13:27:14] src/operator/subgraph/mkldnn/mkldnn_conv_property.cc:138: Start to execute MKLDNN Convolution optimization pass. > [13:27:14] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 147456 bytes with malloc directly > [13:27:14] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 589824 bytes with malloc directly > [13:27:14] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 2359296 bytes with malloc directly > INFO:logger:batch size 64, image/sec: 376.824585 > > SqueezeNet 1.0 > > INFO:logger:batch size = 64 for inference > INFO:logger:rgb_mean = 123.68,116.779,103.939 > INFO:logger:rgb_std = 58.393,57.12,57.375 > INFO:logger:label_name = softmax_label > INFO:logger:Input data shape = (3, 224, 224) > INFO:logger:Dataset for inference: ./data/val_256_q90.rec > [13:33:53] src/io/iter_image_recordio_2.cc:172: ImageRecordIOParser2: ./data/val_256_q90.rec, use 1 threads for decoding.. > INFO:logger:Loading symbol from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/squeezenet1.0-symbol.json > INFO:logger:Loading params from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/squeezenet1.0-0000.params > INFO:logger:Skipping the first 50 batches > INFO:logger:Running model ./model/squeezenet1.0-symbol.json for inference > [13:33:57] src/operator/subgraph/mkldnn/mkldnn_conv_property.cc:138: Start to execute MKLDNN Convolution optimization pass. > [13:33:58] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 71663616 bytes with malloc directly > INFO:logger:Finished inference with 32000 images > INFO:logger:Finished with 118.483729 images per second > INFO:logger:('accuracy', 0.57003125) > INFO:logger:('top_k_accuracy_5', 0.79559375) > INFO:logger:batch size = 64 for inference > INFO:logger:rgb_mean = 123.68,116.779,103.939 > INFO:logger:rgb_std = 58.393,57.12,57.375 > INFO:logger:label_name = softmax_label > INFO:logger:Input data shape = (3, 224, 224) > INFO:logger:Dataset for inference: ./data/val_256_q90.rec > [13:38:30] src/io/iter_image_recordio_2.cc:172: ImageRecordIOParser2: ./data/val_256_q90.rec, use 1 threads for decoding.. > INFO:logger:Loading symbol from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/squeezenet1.0-quantized-5batches-naive-symbol.json > INFO:logger:Loading params from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/squeezenet1.0-quantized-0000.params > INFO:logger:Skipping the first 50 batches > INFO:logger:Running model ./model/squeezenet1.0-quantized-5batches-naive-symbol.json for inference > [13:38:34] src/operator/subgraph/mkldnn/mkldnn_conv_property.cc:138: Start to execute MKLDNN Convolution optimization pass. > [13:38:35] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 512000 bytes with malloc directly > INFO:logger:Finished inference with 32000 images > INFO:logger:Finished with 107.084923 images per second > INFO:logger:('accuracy', 0.5671875) > INFO:logger:('top_k_accuracy_5', 0.79346875) > INFO:logger:batch size = 64 for inference > INFO:logger:rgb_mean = 0,0,0 > INFO:logger:rgb_std = 1,1,1 > INFO:logger:label_name = softmax_label > INFO:logger:Input data shape = (3, 224, 224) > INFO:logger:Running model ./model/squeezenet1.0-symbol.json for inference > INFO:logger:Loading symbol from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/squeezenet1.0-symbol.json > /home/junmingc/xinyu-pr/incubator-mxnet/python/mxnet/module/base_module.py:68: UserWarning: Data provided by label_shapes don't match names specified by label_names ([] vs. ['softmax_label']) > warnings.warn(msg) > [13:43:36] src/operator/subgraph/mkldnn/mkldnn_conv_property.cc:138: Start to execute MKLDNN Convolution optimization pass. > [13:43:36] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 71663616 bytes with malloc directly > INFO:logger:batch size 64, image/sec: 150.544443 > INFO:logger:batch size = 64 for inference > INFO:logger:rgb_mean = 0,0,0 > INFO:logger:rgb_std = 1,1,1 > INFO:logger:label_name = softmax_label > INFO:logger:Input data shape = (3, 224, 224) > INFO:logger:Running model ./model/squeezenet1.0-quantized-5batches-naive-symbol.json for inference > INFO:logger:Loading symbol from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/squeezenet1.0-quantized-5batches-naive-symbol.json > /home/junmingc/xinyu-pr/incubator-mxnet/python/mxnet/module/base_module.py:68: UserWarning: Data provided by label_shapes don't match names specified by label_names ([] vs. ['softmax_label']) > warnings.warn(msg) > [13:47:14] src/operator/subgraph/mkldnn/mkldnn_conv_property.cc:138: Start to execute MKLDNN Convolution optimization pass. > [13:47:14] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 512000 bytes with malloc directly > INFO:logger:batch size 64, image/sec: 135.984801 > > MobileNet 1.0 > > INFO:logger:batch size = 64 for inference > INFO:logger:rgb_mean = 123.68,116.779,103.939 > INFO:logger:rgb_std = 58.393,57.12,57.375 > INFO:logger:label_name = softmax_label > INFO:logger:Input data shape = (3, 224, 224) > INFO:logger:Dataset for inference: ./data/val_256_q90.rec > [13:55:42] src/io/iter_image_recordio_2.cc:172: ImageRecordIOParser2: ./data/val_256_q90.rec, use 1 threads for decoding.. > INFO:logger:Loading symbol from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/mobilenet1.0-symbol.json > INFO:logger:Loading params from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/mobilenet1.0-0000.params > INFO:logger:Skipping the first 50 batches > INFO:logger:Running model ./model/mobilenet1.0-symbol.json for inference > [13:55:46] src/operator/subgraph/mkldnn/mkldnn_conv_property.cc:138: Start to execute MKLDNN Convolution optimization pass. > [13:55:46] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 8192 bytes with malloc directly > [13:55:46] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 32768 bytes with malloc directly > INFO:logger:Finished inference with 32000 images > INFO:logger:Finished with 621.256745 images per second > INFO:logger:('accuracy', 0.72234375) > INFO:logger:('top_k_accuracy_5', 0.90734375) > INFO:logger:batch size = 64 for inference > INFO:logger:rgb_mean = 123.68,116.779,103.939 > INFO:logger:rgb_std = 58.393,57.12,57.375 > INFO:logger:label_name = softmax_label > INFO:logger:Input data shape = (3, 224, 224) > INFO:logger:Dataset for inference: ./data/val_256_q90.rec > [13:56:39] src/io/iter_image_recordio_2.cc:172: ImageRecordIOParser2: ./data/val_256_q90.rec, use 1 threads for decoding.. > INFO:logger:Loading symbol from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/mobilenet1.0-quantized-5batches-naive-symbol.json > INFO:logger:Loading params from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/mobilenet1.0-quantized-0000.params > INFO:logger:Skipping the first 50 batches > INFO:logger:Running model ./model/mobilenet1.0-quantized-5batches-naive-symbol.json for inference > [13:56:43] src/operator/subgraph/mkldnn/mkldnn_conv_property.cc:138: Start to execute MKLDNN Convolution optimization pass. > [13:56:44] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 8192 bytes with malloc directly > INFO:logger:Finished inference with 32000 images > INFO:logger:Finished with 177.447123 images per second > INFO:logger:('accuracy', 0.71921875) > INFO:logger:('top_k_accuracy_5', 0.90553125) > > Inception-V3 > > INFO:logger:batch size = 64 for inference > INFO:logger:rgb_mean = 123.68,116.779,103.939 > INFO:logger:rgb_std = 58.393,57.12,57.375 > INFO:logger:label_name = softmax_label > INFO:logger:Input data shape = (3, 299, 299) > INFO:logger:Dataset for inference: ./data/val_256_q90.rec > [14:07:11] src/io/iter_image_recordio_2.cc:172: ImageRecordIOParser2: ./data/val_256_q90.rec, use 1 threads for decoding.. > INFO:logger:Loading symbol from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/inceptionv3-symbol.json > INFO:logger:Loading params from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/inceptionv3-0000.params > INFO:logger:Skipping the first 50 batches > INFO:logger:Running model ./model/inceptionv3-symbol.json for inference > [14:07:17] src/operator/subgraph/mkldnn/mkldnn_conv_property.cc:138: Start to execute MKLDNN Convolution optimization pass. > [14:07:18] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 36864 bytes with malloc directly > [14:07:18] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 552960 bytes with malloc directly > [14:07:18] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 3981312 bytes with malloc directly > INFO:logger:Finished inference with 32000 images > INFO:logger:Finished with 194.935475 images per second > INFO:logger:('accuracy', 0.77984375) > INFO:logger:('top_k_accuracy_5', 0.93928125) > INFO:logger:batch size = 64 for inference > INFO:logger:rgb_mean = 123.68,116.779,103.939 > INFO:logger:rgb_std = 58.393,57.12,57.375 > INFO:logger:label_name = softmax_label > INFO:logger:Input data shape = (3, 299, 299) > INFO:logger:Dataset for inference: ./data/val_256_q90.rec > [14:10:04] src/io/iter_image_recordio_2.cc:172: ImageRecordIOParser2: ./data/val_256_q90.rec, use 1 threads for decoding.. > INFO:logger:Loading symbol from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/inceptionv3-quantized-5batches-naive-symbol.json > INFO:logger:Loading params from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/inceptionv3-quantized-0000.params > INFO:logger:Skipping the first 50 batches > INFO:logger:Running model ./model/inceptionv3-quantized-5batches-naive-symbol.json for inference > [14:10:10] src/operator/subgraph/mkldnn/mkldnn_conv_property.cc:138: Start to execute MKLDNN Convolution optimization pass. > [14:10:11] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 9216 bytes with malloc directly > [14:10:11] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 138240 bytes with malloc directly > [14:10:11] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 995328 bytes with malloc directly > INFO:logger:Finished inference with 32000 images > INFO:logger:Finished with 102.014292 images per second > INFO:logger:('accuracy', 0.781125) > INFO:logger:('top_k_accuracy_5', 0.93915625) > > ResNet152-V2 > > INFO:logger:batch size = 64 for inference > INFO:logger:rgb_mean = 0,0,0 > INFO:logger:rgb_std = 1,1,1 > INFO:logger:label_name = softmax_label > INFO:logger:Input data shape = (3, 224, 224) > INFO:logger:Dataset for inference: ./data/val_256_q90.rec > [15:27:18] src/io/iter_image_recordio_2.cc:172: ImageRecordIOParser2: ./data/val_256_q90.rec, use 1 threads for decoding.. > INFO:logger:Loading symbol from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/imagenet1k-resnet-152-symbol.json > [15:27:19] src/nnvm/legacy_json_util.cc:209: Loading symbol saved by previous version v0.8.0. Attempting to upgrade... > [15:27:19] src/nnvm/legacy_json_util.cc:217: Symbol successfully upgraded! > INFO:logger:Loading params from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/imagenet1k-resnet-152-0000.params > INFO:logger:Skipping the first 50 batches > INFO:logger:Running model ./model/imagenet1k-resnet-152-symbol.json for inference > [15:27:23] src/operator/subgraph/mkldnn/mkldnn_conv_property.cc:138: Start to execute MKLDNN Convolution optimization pass. > [15:27:24] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 147456 bytes with malloc directly > [15:27:24] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 589824 bytes with malloc directly > [15:27:24] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 2359296 bytes with malloc directly > [15:27:25] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 9437184 bytes with malloc directly > INFO:logger:Finished inference with 32000 images > INFO:logger:Finished with 107.342105 images per second > INFO:logger:('accuracy', 0.7676875) > INFO:logger:('top_k_accuracy_5', 0.93034375) > INFO:logger:batch size = 64 for inference > INFO:logger:rgb_mean = 0,0,0 > INFO:logger:rgb_std = 1,1,1 > INFO:logger:label_name = softmax_label > INFO:logger:Input data shape = (3, 224, 224) > INFO:logger:Dataset for inference: ./data/val_256_q90.rec > [15:32:25] src/io/iter_image_recordio_2.cc:172: ImageRecordIOParser2: ./data/val_256_q90.rec, use 1 threads for decoding.. > INFO:logger:Loading symbol from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/imagenet1k-resnet-152-quantized-5batches-naive-symbol.json > INFO:logger:Loading params from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/imagenet1k-resnet-152-quantized-0000.params > INFO:logger:Skipping the first 50 batches > INFO:logger:Running model ./model/imagenet1k-resnet-152-quantized-5batches-naive-symbol.json for inference > [15:32:31] src/operator/subgraph/mkldnn/mkldnn_conv_property.cc:138: Start to execute MKLDNN Convolution optimization pass. > [15:32:32] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 147456 bytes with malloc directly > [15:32:32] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 589824 bytes with malloc directly > [15:32:33] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 2359296 bytes with malloc directly > INFO:logger:Finished inference with 32000 images > INFO:logger:Finished with 49.981990 images per second > INFO:logger:('accuracy', 0.764875) > INFO:logger:('top_k_accuracy_5', 0.929625) > > > > Inception-BN > > > INFO:logger:batch size = 64 for inference > INFO:logger:rgb_mean = 123.68,116.779,103.939 > INFO:logger:rgb_std = 1,1,1 > INFO:logger:label_name = softmax_label > INFO:logger:Input data shape = (3, 224, 224) > INFO:logger:Dataset for inference: ./data/val_256_q90.rec > [14:35:54] src/io/iter_image_recordio_2.cc:172: ImageRecordIOParser2: ./data/val_256_q90.rec, use 1 threads for decoding.. > INFO:logger:Loading symbol from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/imagenet1k-inception-bn-symbol.json > [14:35:55] src/nnvm/legacy_json_util.cc:209: Loading symbol saved by previous version v0.8.0. Attempting to upgrade... > [14:35:55] src/nnvm/legacy_json_util.cc:217: Symbol successfully upgraded! > INFO:logger:Loading params from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/imagenet1k-inception-bn-0000.params > INFO:logger:Skipping the first 50 batches > INFO:logger:Running model ./model/imagenet1k-inception-bn-symbol.json for inference > [14:35:58] src/operator/subgraph/mkldnn/mkldnn_conv_property.cc:138: Start to execute MKLDNN Convolution optimization pass. > [14:35:58] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 442368 bytes with malloc directly > INFO:logger:Finished inference with 32000 images > INFO:logger:Finished with 445.455068 images per second > INFO:logger:('accuracy', 0.72096875) > INFO:logger:('top_k_accuracy_5', 0.9060625) > INFO:logger:batch size = 64 for inference > INFO:logger:rgb_mean = 123.68,116.779,103.939 > INFO:logger:rgb_std = 1,1,1 > INFO:logger:label_name = softmax_label > INFO:logger:Input data shape = (3, 224, 224) > INFO:logger:Dataset for inference: ./data/val_256_q90.rec > > [14:37:13] src/io/iter_image_recordio_2.cc:172: ImageRecordIOParser2: ./data/val_256_q90.rec, use 1 threads for decoding.. > INFO:logger:Loading symbol from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/imagenet1k-inception-bn-quantized-5batches-naive-symbol.json > INFO:logger:Loading params from file /home/junmingc/xinyu-pr/incubator-mxnet/example/quantization/./model/imagenet1k-inception-bn-quantized-0000.params > INFO:logger:Skipping the first 50 batches > INFO:logger:Running model ./model/imagenet1k-inception-bn-quantized-5batches-naive-symbol.json for inference > [14:37:17] src/operator/subgraph/mkldnn/mkldnn_conv_property.cc:138: Start to execute MKLDNN Convolution optimization pass. > [14:37:17] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 110592 bytes with malloc directly > INFO:logger:Finished inference with 32000 images > INFO:logger:Finished with 504.734005 images per second > INFO:logger:('accuracy', 0.72) > INFO:logger:('top_k_accuracy_5', 0.90534375) > > > > SSD-VGG > > > > INFO:root:Finished inference with 2240 images > INFO:root:Finished with 38.558065 images per second > aeroplane: 0.9057971014492755 > bicycle: 0.8745079058420262 > bird: 0.8407998824780214 > boat: 0.8172609700362614 > bottle: 0.5668336378418553 > bus: 0.8078869544773997 > car: 0.8794030962023759 > cat: 0.9079754601226996 > chair: 0.7888907332306704 > cow: 0.8945315505963835 > diningtable: 0.7868957871396894 > dog: 0.9082568807339451 > horse: 0.9072605247101893 > motorbike: 0.8943175964203456 > person: 0.8056045508318882 > pottedplant: 0.6446982570250784 > sheep: 0.7651375306023592 > sofa: 0.9021571648690294 > train: 0.9090909090909093 > tvmonitor: 0.8749246403233354 > mAP: 0.8341115567011869 > [15:05:39] src/io/iter_image_det_recordio.cc:283: ImageDetRecordIOParser: /home/junmingc/xinyu-pr/incubator-mxnet/example/ssd/data/val.rec, use 32 threads for decoding.. > [15:05:39] src/io/iter_image_det_recordio.cc:340: ImageDetRecordIOParser: /home/junmingc/xinyu-pr/incubator-mxnet/example/ssd/data/val.rec, label padding width: 254 > [15:05:42] src/operator/subgraph/mkldnn/mkldnn_conv_property.cc:138: Start to execute MKLDNN Convolution optimization pass. > [15:05:44] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 36864 bytes with malloc directly > [15:05:52] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 297271296 bytes with malloc directly > [15:05:53] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 628031488 bytes with malloc directly > INFO:root:Finished inference with 2240 images > INFO:root:Finished with 49.863403 images per second > aeroplane: 0.9077238550922763 > bicycle: 0.8582335743673779 > bird: 0.8019871500955977 > boat: 0.8060330180036129 > bottle: 0.5698642432266028 > bus: 0.8507464458004145 > car: 0.8452099241062656 > cat: 0.9085262563523435 > chair: 0.7528436223937933 > cow: 0.8954234786153293 > diningtable: 0.7521532992180973 > dog: 0.9086700336700338 > horse: 0.9011363636363638 > motorbike: 0.8766282736870973 > person: 0.7914271904834184 > pottedplant: 0.6103364637135992 > sheep: 0.7632195285761423 > sofa: 0.9040495867768596 > train: 0.9090909090909093 > tvmonitor: 0.8664679683062037 > mAP: 0.823988559260617
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
