Skip to content

Instantly share code, notes, and snippets.

View f-rumblefish's full-sized avatar

f-rumblefish

4 followers · 1 following
View GitHub Profile
@f-rumblefish
f-rumblefish / object_detection_yolo.py
Last active October 23, 2018 01:58
Object Detection via YOLO
config = 'yolov2.cfg'
model = 'yolov2.weights'
Net_YOLO = cv.dnn.readNetFromDarknet(config, model)
Net_YOLO.setInput(cv.dnn.blobFromImage(img, 1.0/255.0, (416, 416), swapRB = True, crop = False))
Net_YOLO_pred = Net_YOLO.forward()
print(">>> YOLOv2 prediction shape = ", Net_YOLO_pred.shape)
@f-rumblefish
f-rumblefish / performance.csv
Last active September 18, 2018 14:47
CPU GPU
Tensorflow (CPU) PlaidML (Intel GPU) PlaidML (AMD GPU)
Training Time 16969 sec 4314 sec 1841 sec
Evaluation Time 18 sec 21 sec 13 sec
Evaluation Accuacy 67% 69% 68%
@f-rumblefish
f-rumblefish / PlaidML Setup
Created September 18, 2018 10:05
PlaidML Configuration
Experimental Config Devices:
opencl_amd_hainan.0 : Advanced Micro Devices, Inc. Hainan (OpenCL)
opencl_cpu.0 : Intel(R) Corporation CPU (OpenCL)
opencl_intel_intel(r)_hd_graphics_5500.0 : Intel(R) Corporation Intel(R) HD Graphics 5500 (OpenCL)
Using experimental devices can cause poor performance, crashes, and other nastiness.
Enable experimental device support? (y,n)[n]:y
Multiple devices detected (You can override by setting PLAIDML_DEVICE_IDS).
@f-rumblefish
f-rumblefish / Performance_Evaluation.csv
Last active September 18, 2018 10:18
Performance
Tensorflow (CPU) PlaidML (Intel GPU) PlaidML (AMD GPU)
Training Time 16969 sec 4314 sec 1814 sec
Evaluation Time 18 sec 21 sec 13 sec
Evaluation Accuacy 0.67% 0.69% 0.68%
@f-rumblefish
f-rumblefish / Perfect Training
Created September 10, 2018 07:15
Perfect Training
>>> CNN Model Training ...
Train on 2275 samples, validate on 253 samples
Epoch 1/13
2275/2275 [==============================] - 2990s 1s/step - loss: 0.3658 - acc: 0.8958 - val_loss: 1.3321 - val_acc: 0.7628
Epoch 2/13
2275/2275 [==============================] - 3011s 1s/step - loss: 0.1767 - acc: 0.9499 - val_loss: 2.3750 - val_acc: 0.6443
Epoch 3/13
2275/2275 [==============================] - 2926s 1s/step - loss: 0.1461 - acc: 0.9582 - val_loss: 0.8264 - val_acc: 0.7747
Epoch 4/13
2275/2275 [==============================] - 2779s 1s/step - loss: 0.0661 - acc: 0.9811 - val_loss: 0.1774 - val_acc: 0.9447
@f-rumblefish
f-rumblefish / object detection summary.csv
Last active October 22, 2018 08:43
Object Detection Summary
Object Detection Model SSD YOLOv2
Object Classification Model MobileNet Darknet-19
Pre-trained Dataset COCO COCO
Deep Learning Software Platform Tensorflow Darknet
OpenCV ReadNet Function Call readNetFromTensorflow readNetFromDarknet
Config File frozen_inference_graph.pb yolov2.cfg
Weight File ssd_mobilenet_v1_coco_2017_11_17.pbtxt yolov2.weights
OpenCV Forward Function Return (1, 1, 100, 7) (845, 85)
Representation of Bounding Box (left, top, right, bottom) (x_center, y_center, width, height)
Representation of Class Name 90 80
@f-rumblefish
f-rumblefish / music_audio_dataset.csv
Last active July 24, 2018 01:49
Cat/Dog Audio British Birdsong Music Genre Urban Sound
Source Kaggle Kaggle Marsyas UrbanSoundDataset
Total Size 49 MB 633 MB 1 GB 6 GB
Number of Audio Files (total) 277 264 1000 8732
Number of Audio Files (training) - - - 5435*
Number of Audio files (testing) - - - 3297*
Number of Classes 2 88 10 10
Audio Files per Class cat (164) 3 blues (100) air_conditioner (1000)
dog (113) classical (100) children_playing (1000)
country (100) dog_bark (1000)
@f-rumblefish
f-rumblefish / image_dataset.csv
Last active July 15, 2018 13:18
Data Table
MNIST CIFAR Fashion Dog/Cat Shape Fruit Distracted Driver Hand Gesture
Source Keras Keras Keras Home-Made ? ? ? ?
Total Size ? ? ? ? ? ? ? ?
Number of Image Files (total) 70000 60000 70000 ? ? ? ? ?
Number of Image Files (training) 60000 50000 60000 ? ? ? ? ?
Number of Image files (testing) 10000 10000 10000 ? ? ? ? ?
Number of Classes 10 10/100 10 2 ? ? ? ?
Image Files per Class 7000 6000/600 7000 >1000 ? ? ? ?
Size per Image File (max/min) 28x28 32x32 28x28 ? ? ? ? ?
Format of Image File numpy numpy numpy ? ? ? ? ?
@f-rumblefish
f-rumblefish / music_audio_dataset-1.csv
Last active July 24, 2018 01:45
data table
Cat/Dog Audio British Birdsong Heartbeat Music Genre Urban Sound
Source Kaggle Kaggle Kaggle Marsyas UrbanSoundDataset
Total Size 49 MB 633 MB 111 MB 1 GB 6 GB
Number of Audio Files (total) 277 264 ? 1000 8732
Number of Audio Files (training) - - ? - -
Number of Audio files (testing) - - ? - -
Number of Classes 2 88 ? 10 10
Audio Files per Class cat (164) 3 ? blues (100) air_conditioner (1000)
dog (113) classical (100) children_playing (1000)
country (100) dog_bark (1000)
@f-rumblefish
f-rumblefish / cam.py
Last active July 17, 2019 09:29
Class Activation Map
# import Numpy, Scipy, and Matplotlib
import numpy as np
import scipy as sp
import matplotlib.pyplot as plt
# import Keras's functional api
from keras.models import Model
# get the weights from the last layer
gap_weights = model.layers[-1].get_weights()[0]