def getDetections(width, height, class_name, dets, im_name, out_file,
CONF_THRESH, FILE_DELIMITER, __appcfg):
print("getDetections")
row = None
all_bbox = []
inds = np.where(dets[:, -1] >= CONF_THRESH)[0]
print(len(inds))
fileName = Util.getOutFileName(out_file, im_name, ".csv", __appcfg)
with open(fileName, 'a') as f:
if len(inds) == 0:
if __appcfg.SAVE_NULL_RESULTS:
row = Util.getOutFileRow([], class_name, "null", width, height,
FILE_DELIMITER)
f.write(row + '\n')
else:
for i in inds:
bbox = dets[i, :4]
score = dets[i, -1]
# mask_rcnn: getOutFileRow: bbox:: image coordinates
# [ 306 23 1080 1920] => [y1,x1,y2,x2] => [top, left, bottom, right] mapping in Util.getOutFileRow
# faster_rcnn_end2end: getOutFileRow::bbox:
# [643.95715 105.885155 717.3395 177.24414 ] => [left, top, right, bottom] => [x1,y1,x2,y2]
# row = Util.getOutFileRow(bbox, class_name, score, width, height, FILE_DELIMITER)
row = Util.getOutFileRow([bbox[1], bbox[0], bbox[3], bbox[2]],
class_name, score, width, height,
FILE_DELIMITER)
print("row:")
print(row)
## TBD: type conversion mapping
all_bbox.append(row.split(FILE_DELIMITER))
print("Detection Row:" + row)
f.write(row + '\n')
all_rows = {"bbox": all_bbox}
return all_rows
def vis_detections(__appcfg, imgFileName, im, class_names, results):
print("vis_detections")
r = results[0]
# file_name = "viz_{:%Y%m%dT%H%M%S}.png".format(datetime.datetime.now())
# visualize.display_instances(im[..., ::-1], r['rois'], r['masks'], r['class_ids'], class_names, r['scores'])
# file_name = imgFileName+"-viz_{:%Y%m%dT%H%M%S}.png".format(datetime.datetime.now())
# file_name = imgFileName+"-"+__appcfg.ID+"-"+"-viz.png"
fpath = osp.dirname(osp.abspath(imgFileName))
file_name = osp.join(fpath, Util.getVizImageFileName(imgFileName, 'viz', __appcfg ))
print("viz_detections: file_name:viz: {}".format(file_name))
# file_name = osp.join(imgFileName,file_name)
viz = display_instances(file_name, im, r['rois'], r['masks'], r['class_ids'], class_names, r['scores'] )
cv2.imwrite(file_name, viz)
## Color splash
# file_name = imgFileName+"-splash_{:%Y%m%dT%H%M%S}.png".format(datetime.datetime.now())
# file_name = imgFileName+"-"+__appcfg.ID+"-"+"-splash.png"
file_name = osp.join(fpath, Util.getVizImageFileName(imgFileName, 'splash', __appcfg ))
# file_name = osp.join(imgFileName,file_name)
splash = getColorSplash(im, r['masks'])
# skimage.io.imshow(splash)
# skimage.io.show()
cv2.imwrite(file_name, splash)
## get mask
# file_name = imgFileName+"-mask_{:%Y%m%dT%H%M%S}.png".format(datetime.datetime.now())
# file_name = imgFileName+"-"+__appcfg.ID+"-"+"-mask.png"
file_name = osp.join(fpath, Util.getVizImageFileName(imgFileName, 'mask', __appcfg ))
# file_name = osp.join(imgFileName,file_name)
cutmask = getSegmentMask(im[..., ::-1], r['masks'])
## Save output
# skimage.io.imshow(cutmask)
# skimage.io.show()
skimage.io.imsave(file_name, cutmask)
def load(dnn, modelDtls, args):
model = None
loadModel = Util.get(dnn, "loadModel")
if loadModel:
model = loadModel(modelDtls, args)
# loadWeights = Util.get(dnn, "loadWeights")
# loadWeights(modelDtls, model)
# mod = import_module(dnn)
# if mod:
# loadModel = getattr(mod, "loadModel")
# model = loadModel(modelDtls, args)
# print("load: {}".format(mod))
return model
def predict(modelDtls, model, im_name, path, out_file, __appcfg):
print("Inside {}: predict()".format(__file__))
# Load the image
im_name = im_name.split(osp.sep)[-1]
im_file = osp.join(path, im_name)
size_image = osp.getsize(im_file)
print('predict: path, im_name, im_file: {} {}'.format(path, im_name, im_file))
# im = skimage.io.imread(im_file)
im = cv2.imread(im_file)
# print("predict: {}".format(im))
# Run detection
t1 = time.time()
results = model.detect([im], verbose=1)
CLASSES = modelDtls["CLASSES"]
r = results[0]
t2 = time.time()
time_taken = (t2 - t1)
print ('Total time taken in get_detections: %f seconds' %(time_taken))
## OpenCV read the images in BGR format R=0,G=1,B=2
## hence, when plotting with matplotlib specify the order
im = im[:, :, (2, 1, 0)]
dim = im.shape[:2]
height, width = dim[0], dim[1]
FILE_DELIMITER = __appcfg.FILE_DELIMITER
boxes, masks, ids, scores = r['rois'], r['masks'], r['class_ids'], r['scores']
t1 = time.time()
class_names = CLASSES
jsonres = get_detections(im, r['rois'], r['masks'], r['class_ids'], class_names, r['scores'])
t2 = time.time()
time_taken = (t2 - t1)
print ('Total time taken in get_detections: %f seconds' %(time_taken))
jsonres["filename"] = im_name
jsonres["size"] = size_image
via_jsonres = {}
via_jsonres[im_name+str(size_image)] = jsonres
detections = []
res = Util.createResponseForVisionAPI(im_name, FILE_DELIMITER, __appcfg, via_jsonres, detections, __appcfg.API_VISION_BASE_URL)
return res
def exec_prediction(dnn, modelDtls, net, im_names, path, out_file, __appcfg):
print("Inside: Model::exec_prediction")
total = len(im_names)
count = 0
all_rows_for_all_classes = None
print("exec_prediction:: outfile:" + out_file)
print("exec_prediction:: Total images to be predicted:".format(str(total)))
print('~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~')
## DNN specific calls dunamic binding
# mod = import_module(dnn)
# fn = getattr(mod, "predict")
print("dnn: {}".format(dnn))
fn = Util.get(dnn, "predict")
if not fn:
print("fn: no function in the module: TBD exit")
FILE_DELIMITER = __appcfg.FILE_DELIMITER
for im_name in im_names:
## TBD: out_file with timestamp
# fileName = Util.getOutFileName(out_file, im_name, ".csv", __appcfg)
# print("exec_prediction:: fileName: {}".format(fileName))
# # fileName = getCsvFileName(out_file, im_name, __appcfg)
# with open(fileName,'w') as f:
# header = Util.getOutFileHeader(FILE_DELIMITER)
# f.write(header+'\n')
## vis detection output
all_rows_for_all_classes = fn(modelDtls, net, im_name, path, out_file,
__appcfg)
print("exec_prediction::all_rows_for_all_classes: {}".format(
all_rows_for_all_classes))
++count
# if __appcfg.VIS_DETECTIONS:
# Util.vis_detections_from_csvfile(im_name, FILE_DELIMITER, path, out_file, __appcfg)
# else:
# Util.delete_no_detection_csvfile(im_name, FILE_DELIMITER, path, out_file, __appcfg)
print("-----------------------")
return all_rows_for_all_classes
def warmup(dnn, net):
fn = Util.get(dnn, "warmup")
if fn:
fn(net)
def predict_depricated(modelDtls, model, im_name, path, out_file, __appcfg):
print("Inside {}: predict()".format(__file__))
# Load the image
im_file = osp.join(path, im_name)
print('predict: im_name, im_file: {} {}'.format(im_name, im_file))
# im = skimage.io.imread(im_file)
im = cv2.imread(im_file)
print("predict: {}".format(im))
# Run detection
results = model.detect([im], verbose=1)
CLASSES = modelDtls["CLASSES"]
r = results[0]
imgFileName = Util.getOutFileName(out_file, im_name, "", __appcfg)
## Visualize results
vis_detections(__appcfg, imgFileName, im, CLASSES, results)
# jsonres = display_instances(
# fileName, im, r['rois'], r['masks'], r['class_ids'], CLASSES, r['scores']
# )
# # file_name = im_name+"_{:%Y%m%dT%H%M%S}.csv".format(datetime.datetime.now())
# print("fileName: {}".format(fileName))
# with open(fileName,'a') as f:
# for item in jsonres:
# row = ', '.join([str(i) for i in item.values()])
# print("row: {}".format(row))
# f.write(row+"\n")
## OpenCV read the images in BGR format R=0,G=1,B=2
## hence, when plotting with matplotlib specify the order
im = im[:, :, (2, 1, 0)]
dim = im.shape[:2]
height, width = dim[0], dim[1]
FILE_DELIMITER = __appcfg.FILE_DELIMITER
all_rows_for_all_classes = {}
boxes, masks, ids, scores = r['rois'], r['masks'], r['class_ids'], r['scores']
# max_area will save the largest object for all the detection results
max_area = 0
# n_instances saves the amount of all objects
n_instances = boxes.shape[0]
if not n_instances:
print('NO INSTANCES TO DISPLAY')
else:
assert boxes.shape[0] == masks.shape[-1] == ids.shape[0]
for i in range(n_instances):
if not np.any(boxes[i]):
continue
# compute the square of each object
# y1, x1, y2, x2 = boxes[i]
# bbox = boxes[i]
score = scores[i]
y1, x1, y2, x2 = boxes[i]
square = (y2 - y1) * (x2 - x1)
# use label to select person object from all the classes
cls = CLASSES[ids[i]]
max_area = square
mask = masks[:, :, i]
bbox = boxes[i]
# apply mask for the image
# all_rows = getDetections(width, height, cls, bbox, score, im_name, out_file, FILE_DELIMITER, __appcfg)
print("getDetections")
row = None;
all_bbox = []
fileName = Util.getOutFileName(out_file, im_name, ".csv", __appcfg)
with open(fileName,'a') as f:
# row = Util.getOutFileRow([bbox[1],bbox[0],bbox[3],bbox[2]], cls, score, width, height, FILE_DELIMITER)
row = Util.getOutFileRow(bbox, cls, score, width, height, FILE_DELIMITER)
print("row:")
print(row)
## TBD: type conversion mapping
all_bbox.append(row.split(FILE_DELIMITER))
print("Detection Row:"+row)
f.write(row+'\n')
all_rows = {
"bbox":all_bbox
}
if all_bbox and len(all_bbox) > 0:
all_rows_for_all_classes[cls] = all_rows
else:
all_rows_for_all_classes[cls] = None
detections = [
Util.getVizImageFileName(im_name, None, __appcfg )
,Util.getVizImageFileName(im_name, 'viz', __appcfg )
,Util.getVizImageFileName(im_name, 'splash', __appcfg )
,Util.getVizImageFileName(im_name, 'mask', __appcfg )
]
# print("mask_rcnn::detections: {}".format(detections))
res = Util.createResponseForVisionAPI(im_name, FILE_DELIMITER, __appcfg, all_rows_for_all_classes, detections, __appcfg.API_VISION_BASE_URL)
return res
def predict(modelDtls, net, im_name, path, out_file, __appcfg):
print("Inside {}: predict()".format(__file__))
# Load the image
im_file = osp.join(path, im_name)
print('im_name: ' + im_name)
print('im_file: ' + im_file)
im = cv2.imread(im_file)
# Detect all object classes and regress object bounds
timer = Timer()
timer.tic()
scores, boxes = im_detect(net, im)
# print scores,boxes
## OpenCV read the images in BGR format R=0,G=1,B=2
## hence, when plotting with matplotlib specify the order
im = im[:, :, (2, 1, 0)]
modelCfg = modelDtls["config"]
## Ref: https://stackoverflow.com/questions/34768717/matplotlib-unable-to-save-image-in-same-resolution-as-original-image
dim = im.shape[:2]
height, width = dim[0], dim[1]
FILE_DELIMITER = __appcfg.FILE_DELIMITER
timer.toc()
#print(np.amax(scores, axis=1))
#print('Detection took {:.3f}s for {:d} object proposals').format(timer.total_time, boxes.shape[0])
CONF_THRESH = modelCfg.CONF_THRESH
NMS_THRESH = modelCfg.NMS_THRESH
# Visualize detections for each class
CLASSES = modelDtls["CLASSES"]
# print("CLASSES, NMS_THRESH: "+CLASSES+","+NMS_THRESH)
all_rows_for_all_classes = {}
# all_labels = []
labelNames = enumerate(CLASSES)
# print("Label Names: {}").format(CLASSES)
for cls_ind, cls in labelNames:
cls_ind += 1 # because we skipped background
cls_boxes = boxes[:, 4 * cls_ind:4 * (cls_ind + 1)]
cls_scores = scores[:, cls_ind]
dets = np.hstack(
(cls_boxes, cls_scores[:, np.newaxis])).astype(np.float32)
keep = nms(dets, NMS_THRESH)
dets = dets[keep, :]
all_rows = getDetections(width, height, cls, dets, im_name, out_file,
CONF_THRESH, FILE_DELIMITER, __appcfg)
# all_labels.append(cls)
if all_rows and len(all_rows) > 0:
if all_rows["bbox"] and len(all_rows["bbox"]) > 0:
all_rows_for_all_classes[cls] = all_rows
else:
all_rows_for_all_classes[cls] = None
detections = [Util.getVizImageFileName(im_name, None, __appcfg)]
# print("faster_rcnn_end2end::detections: {}".format(detections))
res = Util.createResponseForVisionAPI(im_name, FILE_DELIMITER, __appcfg,
all_rows_for_all_classes, detections,
__appcfg.API_VISION_BASE_URL)
return res