169 lines
6.6 KiB
Python
169 lines
6.6 KiB
Python
from icraft.xir import *
|
|
from icraft.xrt import *
|
|
from icraft.buyibackend import *
|
|
import torch
|
|
import torchvision.transforms as T
|
|
from icraft.host_backend import *
|
|
from PIL import Image
|
|
import numpy as np
|
|
import os
|
|
import matplotlib.pyplot as plt
|
|
|
|
|
|
def dmaInit(device, input_tensor, shape, imk):
|
|
if imk:
|
|
h,w,c=shape[0],shape[1],shape[2]
|
|
demo_reg_base = 0x1000C0000
|
|
uregion_=device.getMemRegion("udma")
|
|
utensor = input_tensor.to(uregion_)#data transfer ps->udma + IMK(udma->pl)
|
|
ImageMakeRddrBase = utensor.data().addr()
|
|
ImageMakeRlen = ((w * h - 1) // (24 // c) + 1) * 3
|
|
ImageMakeLastSft = w * h - (ImageMakeRlen - 3) // 3 * (24 // c)
|
|
device.defaultRegRegion().write(demo_reg_base + 0x4, ImageMakeRddrBase, True)
|
|
device.defaultRegRegion().write(demo_reg_base + 0x8, ImageMakeRlen, True)
|
|
device.defaultRegRegion().write(demo_reg_base + 0xC, ImageMakeLastSft, True)
|
|
device.defaultRegRegion().write(demo_reg_base + 0x10, c, True)
|
|
device.defaultRegRegion().write(demo_reg_base + 0x1C, 1, True)
|
|
device.defaultRegRegion().write(demo_reg_base + 0x20, 0, True)
|
|
# imk start
|
|
device.defaultRegRegion().write(demo_reg_base, 1, True)
|
|
return 0
|
|
|
|
def fpgaOPlist(network):
|
|
# only used in adapt&by stage
|
|
customop_set = set()
|
|
oplist = network.ops
|
|
for op in oplist:
|
|
if "customop" in op.typeKey():
|
|
customop_set.add(op.typeKey())
|
|
return customop_set
|
|
|
|
# colors for visualization
|
|
COLORS = [[0.000, 0.447, 0.741], [0.850, 0.325, 0.098], [0.929, 0.694, 0.125],
|
|
[0.494, 0.184, 0.556], [0.466, 0.674, 0.188], [0.301, 0.745, 0.933]]
|
|
|
|
CLASSES = [
|
|
'N/A', 'person', 'bicycle', 'car', 'motorcycle', 'airplane', 'bus',
|
|
'train', 'truck', 'boat', 'traffic light', 'fire hydrant', 'N/A',
|
|
'stop sign', 'parking meter', 'bench', 'bird', 'cat', 'dog', 'horse',
|
|
'sheep', 'cow', 'elephant', 'bear', 'zebra', 'giraffe', 'N/A', 'backpack',
|
|
'umbrella', 'N/A', 'N/A', 'handbag', 'tie', 'suitcase', 'frisbee', 'skis',
|
|
'snowboard', 'sports ball', 'kite', 'baseball bat', 'baseball glove',
|
|
'skateboard', 'surfboard', 'tennis racket', 'bottle', 'N/A', 'wine glass',
|
|
'cup', 'fork', 'knife', 'spoon', 'bowl', 'banana', 'apple', 'sandwich',
|
|
'orange', 'broccoli', 'carrot', 'hot dog', 'pizza', 'donut', 'cake',
|
|
'chair', 'couch', 'potted plant', 'bed', 'N/A', 'dining table', 'N/A',
|
|
'N/A', 'toilet', 'N/A', 'tv', 'laptop', 'mouse', 'remote', 'keyboard',
|
|
'cell phone', 'microwave', 'oven', 'toaster', 'sink', 'refrigerator', 'N/A',
|
|
'book', 'clock', 'vase', 'scissors', 'teddy bear', 'hair drier',
|
|
'toothbrush'
|
|
]
|
|
transform = T.Compose([
|
|
T.Resize([640,640]),
|
|
|
|
])
|
|
def dmaInit(shape,input_tensor,device ):
|
|
h,w,c=shape[0],shape[1],shape[2]
|
|
demo_reg_base = 0x1000C0000
|
|
uregion_=device.getMemRegion("udma")
|
|
utensor = input_tensor.to(uregion_)#data transfer ps->udma + IMK(udma->pl)
|
|
ImageMakeRddrBase = utensor.data().addr()
|
|
ImageMakeRlen = ((w * h - 1) // (24 // c) + 1) * 3
|
|
ImageMakeLastSft = w * h - (ImageMakeRlen - 3) // 3 * (24 // c)
|
|
device.defaultRegRegion().write(demo_reg_base + 0x4, ImageMakeRddrBase, True)
|
|
device.defaultRegRegion().write(demo_reg_base + 0x8, ImageMakeRlen, True)
|
|
device.defaultRegRegion().write(demo_reg_base + 0xC, ImageMakeLastSft, True)
|
|
device.defaultRegRegion().write(demo_reg_base + 0x10, c, True)
|
|
device.defaultRegRegion().write(demo_reg_base + 0x1C, 1, True)
|
|
device.defaultRegRegion().write(demo_reg_base + 0x20, 0, True)
|
|
# imk start
|
|
device.defaultRegRegion().write(demo_reg_base, 1, True)
|
|
return 0
|
|
|
|
def box_cxcywh_to_xyxy(x):
|
|
x_c, y_c, w, h = x.unbind(1)
|
|
b = [(x_c - 0.5 * w), (y_c - 0.5 * h),
|
|
(x_c + 0.5 * w), (y_c + 0.5 * h)]
|
|
return torch.stack(b, dim=1)
|
|
def rescale_bboxes(out_bbox, size):
|
|
img_w, img_h = size
|
|
b = box_cxcywh_to_xyxy(out_bbox)
|
|
b = b * torch.tensor([img_w, img_h, img_w, img_h], dtype=torch.float32)
|
|
return b
|
|
|
|
|
|
def plot_results(pil_img, prob, boxes):
|
|
plt.figure(figsize=(16,10))
|
|
plt.imshow(pil_img)
|
|
ax = plt.gca()
|
|
for p, (xmin, ymin, xmax, ymax), c in zip(prob, boxes.tolist(), COLORS * 100):
|
|
ax.add_patch(plt.Rectangle((xmin, ymin), xmax - xmin, ymax - ymin,
|
|
fill=False, color=c, linewidth=3))
|
|
cl = p.argmax()
|
|
text = f'{CLASSES[cl]}: {p[cl]:0.2f}'
|
|
ax.text(xmin, ymin, text, fontsize=15,
|
|
bbox=dict(facecolor='yellow', alpha=0.5))
|
|
plt.axis('off')
|
|
plt.show()
|
|
|
|
|
|
|
|
run_parse = False
|
|
JSON_FILE = '../3_deploy/modelzoo/detr/imodel/8/detr_640x640_optimized.json'
|
|
RAW_FILE = '../3_deploy/modelzoo/detr/imodel/8/detr_640x640_optimized.raw'
|
|
IMG_PATH = '../2_compile/qtset/detr/000000000632.jpg'
|
|
|
|
network = Network.CreateFromJsonFile(JSON_FILE)
|
|
network.loadParamsFromFile(RAW_FILE)
|
|
customop_set=fpgaOPlist(network)
|
|
print(customop_set)
|
|
IMK = True if "customop::ImageMakeNode" in customop_set else False
|
|
print('IMK =',IMK)
|
|
device = HostDevice.Default()
|
|
|
|
sess = Session.Create([ HostBackend ], network.view(0), [HostDevice.Default()])
|
|
|
|
|
|
im = Image.open(IMG_PATH)
|
|
img = np.array(transform(im)).reshape(1,640,640,3).astype(np.float32).copy()
|
|
# img2 = img2.permute(0, 2, 3, 1).cpu().detach().numpy().astype(np.float32).copy()
|
|
if IMK:
|
|
img= img.astype(np.uint8)
|
|
else:
|
|
img= img.astype(np.float32)
|
|
|
|
tensor = Tensor(img, Layout("NHWC"))
|
|
|
|
sess.enableTimeProfile(True)
|
|
|
|
sess.apply()
|
|
|
|
output_tensors = sess.forward([tensor])
|
|
|
|
# print(output_tensors[0])
|
|
# print(output_tensors[1])
|
|
device.reset(1)
|
|
result = sess.timeProfileResults()
|
|
#[总时间，传输时间，硬件时间，余下时间]
|
|
# time = np.array(list(result.values()))
|
|
# total_hardtime = np.sum(time[:,2])
|
|
# print("Total Time: ",np.sum(time[:,0]),"ms")
|
|
# print("Travel Time: ",np.sum(time[:,1]),"ms")
|
|
# print("Hard Time: ",np.sum(time[:,2]),"ms")
|
|
# print("Res Time: ",np.sum(time[:,3]),"ms")
|
|
|
|
outputs_class_ = np.reshape(output_tensors[0], (1,100,92))
|
|
outputs_class_icraft = torch.tensor(outputs_class_).reshape(1,100,92).contiguous()
|
|
outputs_coord_ = np.reshape(output_tensors[1], (1,100,4))
|
|
outputs_coord_icraft = torch.tensor(outputs_coord_).reshape(1,100,4).contiguous()
|
|
outputs_class = outputs_class_icraft
|
|
outputs_coord = outputs_coord_icraft
|
|
probas = outputs_class.softmax(-1)[0, :, :-1]
|
|
keep = probas.max(-1).values > 0.7
|
|
|
|
# convert boxes from [0; 1] to image scales
|
|
bboxes_scaled = rescale_bboxes(outputs_coord[0, keep], (640, 483))
|
|
|
|
plot_results(im, probas[keep], bboxes_scaled)
|
|
|