Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

Sign up
Appearance settings

How to train using custom model #178

Unanswered
LeMosquitar asked this question in Q&A
Discussion options

Alright, I have seen the threads in issues before coming here. And, thank you everyone for your efforts.
Here is the thing, I consulted https://yolo-docs.readthedocs.io/en/latest/1_tutorials/0_allIn1.html. Followed the folder structure and run the training with lazy.py (use_wandb=False). However, I keep getting the output below, things do not seem to progress, there was even a time I waited for an hour. I used Google Colab for T4 GPU, same results, I used my own PC, CPU, same results. I also tried to export and used dataset format YOLOv9, same result, stuck in the first progress. Any help is appreciated on how to train. Thank you.

Command:

python yolo\lazy.py task=train dataset=v9-aug use_wandb=False device=cpu task.data.batch_size=12 name=t002

config.py:

hydra:
 run:
 dir: runs
name: v9-dev
defaults:
 - _self_
 - task: train
 - dataset: v9-aug
 - model: v9-c
 - general

I am using COCO export format of my dataset from ROBOFLOW, I have the following yaml file in yolo/config/dataset/v9-aug.yaml:

path: C:\Users\LM\Downloads\v9-v1_aug.coco
train: C:\Users\LM\Downloads\v9-v1_aug.coco\images\train
validation: C:\Users\LM\Downloads\v9-v1_aug.coco\images\validation
test: C:\Users\LM\Downloads\v9-v1_aug.coco\images\test
class_num: 3
class_list: ['A', 'B', 'C']

Output is:

 INFO 有 Found stride of model [8, 16, 32] 
 INFO ✅ Success load loss function 
┏━━━┳━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━┳━━━━━━━┓
┃ ┃ Name ┃ Type ┃ Params ┃ Mode ┃
┡━━━╇━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━╇━━━━━━━┩
│ 0 │ model │ YOLO │ 51.0 M │ train │
│ 1 │ metric │ MeanAveragePrecision │ 0 │ train │
│ 2 │ ema │ YOLO │ 51.0 M │ train │
└───┴────────┴──────────────────────┴────────┴───────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┓
┃ Attributes ┃ Value ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━┩
│ Trainable params │ 51.0 M │
│ Non-trainable params │ 51.0 M │
│ Total params │ 102 M │
│ Total estimated model params size (MB) │ 408 │
│ Modules in train mode │ 2443 │
│ Modules in eval mode │ 0 │
└────────────────────────────────────────┴────────────┘

I also have the following output log on runs/train:

:chart_with_upwards_trend: Enable Model EMA
:tractor: Building YOLO
 :building_construction: Building backbone
 :building_construction: Building neck
 :building_construction: Building head
 :building_construction: Building detection
 :building_construction: Building auxiliary
:warning: Weight Mismatch for key: 22.heads.0.class_conv
:warning: Weight Mismatch for key: 38.heads.0.class_conv
:warning: Weight Mismatch for key: 22.heads.2.class_conv
:warning: Weight Mismatch for key: 22.heads.1.class_conv
:warning: Weight Mismatch for key: 38.heads.1.class_conv
:warning: Weight Mismatch for key: 38.heads.2.class_conv
:white_check_mark: Success load model & weight
:package: Loaded C:\Users\LM\Downloads\v9-v1_aug.coco\images\validation cache
:package: Loaded C:\Users\LM\Downloads\v9-v1_aug.coco\images\train cache
:japanese_not_free_of_charge_button: Found stride of model [8, 16, 32]
:white_check_mark: Success load loss function
You must be logged in to vote

Replies: 2 comments 1 reply

Comment options

Hello sir, have you found a solution to see the progress of learning?

You must be logged in to vote
0 replies
Comment options

I think this is bug
you can try
python yolo\lazy.py task=train dataset=v9-aug device=gpu task.data.batch_size=12 name=t002

wandb=true,then when ask

INFO 🌐 (1) Create a W&B account logging_utils.py:268
INFO 🌐 (2) Use an existing W&B account logging_utils.py:268
INFO 🌐 (3) Don't visualize my results logging_utils.py:268

select (3)

You must be logged in to vote
1 reply
Comment options

device=cuda or device=gpu? I haven't seen the latter thus far.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

AltStyle によって変換されたページ (->オリジナル) /