from __future__ import division import time import torch import torch.nn as nn from torch.autograd import Variable import numpy as np import cv2 from yolo.utils import * import argparse import os import os.path as osp from yolo.darknet import Darknet # from preprocess import prep_image, inp_to_image import pandas as pd import random import pickle as pkl import itertools import os import base64 from PIL import Image from io import BytesIO class yolo_model(): batch_size = int(1) confidence = float(0.5) nms_thesh = float(0.4) reso = 416 start = 0 CUDA = torch.cuda.is_available() num_classes = 80 def __init__(self): self.classes = load_classes( os.path.join( 'yolo' , 'data', 'coco.names' ) ) # self.colors = pkl.load( get_data_s3( "pallete" ) ) # Set up the neural network self.model = Darknet( os.path.join( 'yolo' , 'yolov3-tiny.cfg' ) ) self.model.load_weights( os.path.join( 'yolo' , 'yolov3-tiny.weights' ) ) print(' [*] Model Loaded Successfuly') # set model resolution self.model.net_info["height"] = self.reso self.inp_dim = int(self.model.net_info["height"]) assert self.inp_dim % 32 == 0 assert self.inp_dim > 32 # If there's a GPU availible, put the model on GPU if self.CUDA: self.model.cuda() # Set the model in evaluation mode self.model.eval() def write( self , x , batches , results , colors=[] ): c1 = tuple(x[1:3].int()) c2 = tuple(x[3:5].int()) img = results[int(x[0])] print( 'img' , int( x[0] ) ) print( 'cls' , int( x[-1] ) ) cls = int(x[-1]) label = "{0}".format(self.classes[cls]) color = random.choice(colors) cv2.rectangle(img, c1, c2,color, 1) t_size = cv2.getTextSize(label, cv2.FONT_HERSHEY_PLAIN, 1 , 1)[0] c2 = c1[0] + t_size[0] + 3, c1[1] + t_size[1] + 4 cv2.rectangle(img, c1, c2,color, -1) cv2.putText(img, label, (c1[0], c1[1] + t_size[1] + 4), cv2.FONT_HERSHEY_PLAIN, 1, [225,255,255], 1) return img def img_to_base64_str(self,img): buffered = BytesIO() img.save(buffered, format="PNG") buffered.seek(0) img_byte = buffered.getvalue() img_str = "data:image/png;base64," + base64.b64encode(img_byte).decode() return img_str def predict( self , image ): imlist = [] imlist.append( image ) batches = list( map( prep_image_org , imlist , [ self.inp_dim for x in range( len(imlist) ) ] ) ) im_batches = [x[0] for x in batches] orig_ims = [x[1] for x in batches] im_dim_list = [x[2] for x in batches] print( 'im_dim_list : ' , im_dim_list ) im_dim_list = torch.FloatTensor(im_dim_list).repeat(1,2) if self.CUDA: im_dim_list = im_dim_list.cuda() print('im_batches' , len(im_batches)) batch = im_batches[0] if self.CUDA: batch = batch.cuda() #Apply offsets to the result predictions #Tranform the predictions as described in the YOLO paper #flatten the prediction vector # B x (bbox cord x no. of anchors) x grid_w x grid_h --> B x bbox x (all the boxes) # Put every proposed box as a row. with torch.no_grad(): prediction = self.model(Variable(batch), self.CUDA) # prediction = prediction[:,scale_indices] #get the boxes with object confidence > threshold #Convert the cordinates to absolute coordinates #perform NMS on these boxes, and save the results #I could have done NMS and saving seperately to have a better abstraction #But both these operations require looping, hence #clubbing these ops in one loop instead of two. #loops are slower than vectorised operations. prediction = write_results(prediction, self.confidence, self.num_classes, nms = True, nms_conf = self.nms_thesh) end = time.time() # print(end - start) # prediction[:,0] += i*batch_size output = prediction # 1, 1, 1 # print( 'enumerate : ' , batch_size , len(imlist) , min( batch_size , len(imlist) ) ) for im_num, image in enumerate( imlist ): im_id = im_num objs = [self.classes[int(x[-1])] for x in output if int(x[0]) == im_id] # print("{0:20s} predicted in {1:6.3f} seconds".format(image.split("/")[-1], (end - self.start)/self.batch_size)) print("{0:20s} {1:s}".format("Objects Detected:", " ".join(objs))) print("----------------------------------------------------------") im_dim_list = torch.index_select(im_dim_list, 0, output[:,0].long()) scaling_factor = torch.min(self.inp_dim/im_dim_list,1)[0].view(-1,1) output[:,[1,3]] -= (self.inp_dim - scaling_factor*im_dim_list[:,0].view(-1,1))/2 output[:,[2,4]] -= (self.inp_dim - scaling_factor*im_dim_list[:,1].view(-1,1))/2 output[:,1:5] /= scaling_factor for i in range(output.shape[0]): output[i, [1,3]] = torch.clamp(output[i, [1,3]], 0.0, im_dim_list[i,0]) output[i, [2,4]] = torch.clamp(output[i, [2,4]], 0.0, im_dim_list[i,1]) colors = pkl.load( open( "yolo/pallete", "rb") ) list(map(lambda x: self.write( x , im_batches , orig_ims , colors=colors ) , output ) ) print('orig_ims : shape ',orig_ims[0].shape) # print('orig_ims : ',orig_ims[0]) output_image = Image.fromarray(orig_ims[0]) img_str = self.img_to_base64_str(output_image) # im_bytes = orig_ims[0].tobytes() # im_b64 = base64.b64encode(im_bytes) # im_b64 = im_b64.decode('utf-8') # print( 'im_b64' , im_b64 ) payload = dict({ 'image' : img_str , 'objects' : objs }) return payload,output_image