Article Outline
Python pil example 'extract fish from eventmeasure'
Functions in program:
def draw_box(image, box, color, thickness=2):def find_fish_bounds(lx0, ly0, lx1, ly1):def find_fish_bounds1(lx0, ly0, lx1, ly1):
Modules used in program:
import osimport cv2import numpy as npimport pandas
python extract fish from eventmeasure
Python pil example: extract fish from eventmeasure
import pandas
import numpy as np
import cv2
import os
from PIL import Image
def find_fish_bounds1(lx0, ly0, lx1, ly1):
original_lx0 = lx0
original_lx1 = lx1
original_ly0 = ly0
original_ly1 = ly1
print("Orig", lx0, ly0, lx1, ly1)
# if fish is measured right to left, we need to swap the xys so we can draw a bounding box consistently
lx0 = original_lx0 if original_lx0 < original_lx1 else original_lx1
lx1 = original_lx0 if original_lx0 > original_lx1 else original_lx1
ly0 = original_ly0 if original_ly0 < original_ly1 else original_ly0
ly1 = original_ly0 if original_ly0 > original_ly1 else original_ly0
# adjust the xys to crop a bounding box
width = lx1 - lx0
quarter_width = width / 4
if not ((ly1-ly0) > (quarter_width*2)):
ly0 -= quarter_width
ly1 += quarter_width
#if lx0 == lx1 or ly0 == ly1:
# continue
if lx0 >= lx1 or ly0 >= ly1:
print("Adju", lx0, ly0, lx1, ly1)
print("False")
#exit(0)
return int(lx0), int(ly0), int(lx1), int(ly1)
def find_fish_bounds(lx0, ly0, lx1, ly1):
original_lx0 = lx0
original_lx1 = lx1
original_ly0 = ly0
original_ly1 = ly1
#print("Orig", lx0, ly0, lx1, ly1)
# if fish is measured right to left, we need to swap the xys so we can draw a bounding box consistently
lx0 = original_lx0 if original_lx0 < original_lx1 else original_lx1
lx1 = original_lx0 if original_lx0 > original_lx1 else original_lx1
ly0 = original_ly0 if original_ly0 < original_ly1 else original_ly1
ly1 = original_ly0 if original_ly0 > original_ly1 else original_ly1
midy = ly0 + ((ly1-ly0)/2)
diffy = ly1 - ly0
# adjust the xys to crop a bounding box
width = lx1 - lx0
quarter_width = width / 3
ly0_tmp = midy - quarter_width
ly1_tmp = midy + quarter_width
if ((diffy) < (ly1_tmp-ly0_tmp)):
ly0 = midy - quarter_width
ly1 = midy + quarter_width
if ly0 < 0:
ly0=0
#if lx0 == lx1 or ly0 == ly1:
# continue
if lx0 >= lx1 or ly0 >= ly1:
print("Adju", lx0, ly0, lx1, ly1)
print("False")
#exit(0)
return int(lx0), int(ly0), int(lx1), int(ly1)
def draw_box(image, box, color, thickness=2):
""" Draws a box on an image with a given color.
# Arguments
image : The image to draw on.
box : A list of 4 elements (x1, y1, x2, y2).
color : The color of the box.
thickness : The thickness of the lines to draw a box with.
"""
b = np.array(box).astype(int)
cv2.rectangle(image, (b[0], b[1]), (b[2], b[3]), color, thickness, cv2.LINE_AA)
if __name__ == "__main__":
em_lengths_file = "/path/file_Lengths.txt"
em_image_pt_pair_file = "/path/file_ImagePtPair.txt"
VIDEO_BASE_DIR = "/path/"
FRAMES_OUTPUT_DIR = "/path/"
df_em_lengths = pandas.read_csv(em_lengths_file, sep='\t', lineterminator='\r')
df_em_pt_pair = pandas.read_csv(em_image_pt_pair_file, sep='\t', lineterminator='\r')
df = pandas.merge(df_em_lengths, df_em_pt_pair, on=['OpCode', 'ImagePtPair'])
df["Index"] = pandas.to_numeric(df["Index"])
df["FrameLeft"] = pandas.to_numeric(df["FrameLeft"])
op_codes = df.OpCode.unique()
point_pairs = df.ImagePtPair.unique()
filenames = df.FilenameLeft.unique()
'''
species_list = df.Species.unique()
genus_list = df.Genus.unique()
family_list = df.Family.unique()
def write_list(filename, the_list):
with open(filename, "w") as text_file:
for index, name in enumerate(the_list):
text_file.write(str(name) + "," + str(index) + "\n")
write_list("species_ids.csv", species_list)
write_list("genus_ids.csv", genus_list)
write_list("family_ids.csv", family_list)
exit(0)
'''
'''
for filename in filenames:
exists = os.path.exists(os.path.join(VIDEO_BASE_DIR, filename))
print(os.path.join(VIDEO_BASE_DIR, filename), exists)
'''
species_df = pandas.DataFrame(columns=['filename', 'x0', 'y0', 'x1', 'y1', 'label'])
genus_df = pandas.DataFrame(columns=['filename', 'x0', 'y0', 'x1', 'y1', 'label'])
family_df = pandas.DataFrame(columns=['filename', 'x0', 'y0', 'x1', 'y1', 'label'])
count = 0
for code in op_codes:
for point in point_pairs:
if df[(df.OpCode == code) & (df.ImagePtPair == point)].shape[0] == 0:
continue
filename = df[(df.OpCode == code) & (df.ImagePtPair == point)]["FilenameLeft"].values[0]
frame = df[(df.OpCode == code) & (df.ImagePtPair == point)]["FrameLeft"].values[0]
lx0 = df[(df.OpCode == code) & (df.ImagePtPair == point) & (df.Index == 0)]["Lx"].values[0]
ly0 = df[(df.OpCode == code) & (df.ImagePtPair == point) & (df.Index == 0)]["Ly"].values[0]
lx1 = df[(df.OpCode == code) & (df.ImagePtPair == point) & (df.Index == 1)]["Lx"].values[0]
ly1 = df[(df.OpCode == code) & (df.ImagePtPair == point) & (df.Index == 1)]["Ly"].values[0]
lx_0 = int(lx0)
ly_0 = int(ly0)
lx_1 = int(lx1)
ly_1 = int(ly1)
midx = int(lx0 + ((lx1 - lx0) / 2))
midy = int(ly0 + ((ly1 - ly0) / 2))
frame = int(frame)
species = df[(df.OpCode == code) & (df.ImagePtPair == point)]["Species"].values[0]
genus = df[(df.OpCode == code) & (df.ImagePtPair == point)]["Genus"].values[0]
family = df[(df.OpCode == code) & (df.ImagePtPair == point)]["Family"].values[0]
#if lx0 > lx1:
# continue
lx0, ly0, lx1, ly1 = find_fish_bounds(lx0, ly0, lx1, ly1)
if lx0 >= lx1 or ly0 >= ly1:
print("skipping", filename, frame, lx0, ly0, lx1, ly1, family, genus, species)
continue
#print(filename, frame, lx0, ly0, lx1, ly1, family, genus, species)
video = os.path.join(VIDEO_BASE_DIR, filename)
if not os.path.exists(video):
print("continuing")
continue
output_frame_path = os.path.join(FRAMES_OUTPUT_DIR, filename + "." + str(int(frame)) + ".png")
species_df = species_df.append(
{"filename": output_frame_path, "x0": lx0, "y0": ly0, "x1": lx1, "y1": ly1, "label": species},
ignore_index=True)
genus_df = genus_df.append(
{"filename": output_frame_path, "x0": lx0, "y0": ly0, "x1": lx1, "y1": ly1, "label": genus},
ignore_index=True)
family_df = family_df.append(
{"filename": output_frame_path, "x0": lx0, "y0": ly0, "x1": lx1, "y1": ly1, "label": family},
ignore_index=True)
# if os.path.exists(output_frame_path):
# continue
cap = cv2.VideoCapture(video) # video_name is the video being called
#amount_of_frames = cap.get(cv2.CAP_PROP_FRAME_COUNT)
#print(amount_of_frames)
cap.set(cv2.CAP_PROP_POS_FRAMES, int(frame)) # Where frame_no is the frame you want
ret, frame = cap.read() # Read the frame
pil_image = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
pil_image = Image.fromarray(pil_image)
pil_image.save(output_frame_path, "PNG")
draw_box(frame, (lx0, ly0, lx1, ly1), (0, 255, 0))
cv2.circle(frame, (lx_0, ly_0), 3, (0, 0, 255), 2, cv2.LINE_AA)
cv2.circle(frame, (lx_1, ly_1), 3, (0, 0, 255), 2, cv2.LINE_AA)
cv2.circle(frame, (midx, midy), 3, (0, 0, 255), 2, cv2.LINE_AA)
#image_utils.draw_bounding_box(frame, lx_0, ly_0, lx_1, ly_1, "yellow")
#frame.show()
#time.sleep(8)
cv2.imshow('window_name', frame)
if cv2.waitKey() == ord('q'):
exit(0)
print(df[(df.OpCode == code) & (df.ImagePtPair == point) & (df.Index == 0)]["Lx"].values[0])
print(df[(df.OpCode == code) & (df.ImagePtPair == point) & (df.Index == 0)]["Ly"].values[0])
print(df[(df.OpCode == code) & (df.ImagePtPair == point) & (df.Index == 0)]["Rx"].values[0])
print(df[(df.OpCode == code) & (df.ImagePtPair == point) & (df.Index == 0)]["Ry"].values[0])
print(df[(df.OpCode == code) & (df.ImagePtPair == point) & (df.Index == 1)]["Lx"].values[0])
print(df[(df.OpCode == code) & (df.ImagePtPair == point) & (df.Index == 1)]["Ly"].values[0])
print(df[(df.OpCode == code) & (df.ImagePtPair == point) & (df.Index == 1)]["Rx"].values[0])
print(df[(df.OpCode == code) & (df.ImagePtPair == point) & (df.Index == 1)]["Ry"].values[0])
#print(df[(df.OpCode == code) & (df.Index == 0)]["Ly"][0])
species_df.to_csv("./outputs/species_train.csv")
family_df.to_csv("./outputs/family_train.csv")
genus_df.to_csv("./outputs/genus_train.csv")
Python links
- Learn Python: https://pythonbasics.org/
- Python Tutorial: https://pythonprogramminglanguage.com