search

ultralytics / yolov5

YOLOv5 🚀 in PyTorch > ONNX > CoreML > TFLite
https://docs.ultralytics.com
GNU Affero General Public License v3.0
49.68k stars 16.11k forks source link

background color of image and other causes? #13298

Open pratikshac15 opened 1 week ago

pratikshac15 commented 1 week ago

Search before asking

Question

Hello, @glenn-jocher I am training my image samples that contain particles on a white background, but I am testing the model with images that have a black background and completely different dataset. What could be the reasons why the model is unable to detect the particles on the black background? Even when it does detect something, it ends up detecting the entire image or a large portion of the black area.

Attached files are some training images.. LUCID_ATX081S-C_221200016__20240716123434318_image96_jpg rf ffd7ca0407b5c46ebcb2d09fadaecc9b LUCID_ATX081S-C_221200016__20240716123859782_image114_jpg rf a6c72d23e8a6556c9198b7b92fdee914 LUCID_ATX081S-C_221200016__20240716124824500_image143_jpg rf 132c1d4da33a57c1a027c106bf96da25 LUCID_ATX081S-C_221200016__20240716124843398_image145_jpg rf 9e3244418077155e74ff893345323938 LUCID_ATX081S-C_221200016__20240716131625093_image226_jpg rf 4f8ceb738680a20705191d01d659ce99 LUCID_ATX081S-C_221200016__20240716142839990_image245_jpg rf 3cd5edd76a681cf8fd41f157ddf61194

Test images 0509-0001 0509-0002 0509-0006 0509-0013 0509-0027 0510-0033 ..

Additional

No response

glenn-jocher commented 1 week ago

@pratikshac15 the model's difficulty in detecting particles on a black background may be due to a lack of diverse training data. Ensure your dataset includes images with varied backgrounds to improve generalization. Also, verify label accuracy and consider retraining with a more representative dataset. For further guidance, check our tips for best training results in the documentation.