search

lucasjinreal / yolov7_d2

🔥🔥🔥🔥 (Earlier YOLOv7 not official one) YOLO with Transformers and Instance Segmentation, with TensorRT acceleration! 🔥🔥🔥
GNU General Public License v3.0
3.13k stars 483 forks source link

How to perform inference for instance segmentation trained on custom data? #204

Closed ankita-chatterjee closed 1 year ago

ankita-chatterjee commented 1 year ago

I tried using my_demo.py using this command !python my_demo.py --config-file output/wall_yolomask1/config.yaml --input dataset/hyper_floorplan_datasets_old/tests/floorplan_68_solid_r90_flipn.jpg --output output/detected/1.jpg --opts MODEL.WEIGHTS output/wall_yolomask1/model_final.pth

I am getting detection images with coco dataset class names not my dataset class names

could you please help me?

lucasjinreal commented 1 year ago

yaml didn't contains any information about your dataset class names (inference will not read your annotation data as well).

You can manually edit demo.py in visualize sending your classes names to visualization func.

ankita-chatterjee commented 1 year ago

Can you tell me the line number in the demo.py. Where do I have to write my class name?

On Wed, Nov 2, 2022 at 8:00 AM JinTian @.***> wrote:

yaml didn't contains any information about your dataset class names (inference will not read your annotation data as well).

You can manually edit demo.py in visualize sending your classes names to visualization func.

— Reply to this email directly, view it on GitHub https://github.com/jinfagang/yolov7_d2/issues/204#issuecomment-1299478161, or unsubscribe https://github.com/notifications/unsubscribe-auth/ANI3YMR6GZ43GP5ICOLBF2TWGHG27ANCNFSM6AAAAAARUKKKA4 . You are receiving this because you authored the thread.Message ID: @.***>

brmarkus commented 1 year ago

You might need to check "https://github.com/jinfagang/yolov7_d2/blob/main/demo.py#L139" with those calls of vis_bitmasks_with_classes().

According to the dependency "alfred", see "https://github.com/jinfagang/alfred/blob/main/alfred/vis/image/mask.py#L313", if no detected-class-name is given, then the original coco_label_map_list will be used.