Getting the optimized model takes forever

[eduardo-lago-aguilar]

Here is my code based on the demo to get the resnet18_baseline_att_224x224_A_epoch_249_trt.pth optimized model:

import json
import trt_pose.coco

with open('human_pose.json', 'r') as f:
    human_pose = json.load(f)

topology = trt_pose.coco.coco_category_to_topology(human_pose)

import trt_pose.models

num_parts = len(human_pose['keypoints'])
num_links = len(human_pose['skeleton'])

model = trt_pose.models.resnet18_baseline_att(num_parts, 2 * num_links).cuda().eval()

print(num_parts)
print(num_links)

import torch

MODEL_WEIGHTS = 'resnet18_baseline_att_224x224_A_epoch_249.pth'

model.load_state_dict(torch.load(MODEL_WEIGHTS))


WIDTH = 224
HEIGHT = 224

data = torch.zeros((1, 3, HEIGHT, WIDTH)).cuda()


import torch2trt

model_trt = torch2trt.torch2trt(model, [data], fp16_mode=True, max_workspace_size=1<<25)

OPTIMIZED_MODEL = 'resnet18_baseline_att_224x224_A_epoch_249_trt.pth'

torch.save(model_trt.state_dict(), OPTIMIZED_MODEL)

it has been running for more than 2 hours already, and the Jetson Nano is not responsive at all.

Is it the expected behavior? How long does it usually take to generate it?

[eduardo-lago-aguilar]

After long wait I got this:

python3 humanpose.py
18
21
[1388.798163] python3 invoked com-killer: gfp_nesk=0x24082c2(GFP KERNELI_GFP_HIGHMEMI_GFP_NOHARNI_GFP ZERO), nodemask=0, order=0, oom score adJ=0 
[1388.812530 ] Out of memory: Kill process 7187 (python3) score 536 or sacrifice child 
[1388.820457 ] KIlled process 7187 (python3) total-vm: 12791888KB, anon-rss:1799072KB, file-rss:367720KB, shmen-rss:0kB KIlled

[gohjiayi]

Hi, it took me ~5 mins to run on a Jetson Nano. I ran the exact same code on trt_pose/tasks/human_pose/live_demo.ipynb. Looks like your Nano ran out of memory, perhaps consider adding swap space to your system. I have followed this tutorial in the past.

[manish33scss]

i have 8gb of swap, and it takes about 5-8 min.