Open panagiotamoraiti opened 5 months ago
Hi @panagiotamoraiti,
You don't have to necessarily use an RTX 3090 GPU. However, you have to keep in mind that the 3090 comes with 24 GB of vRAM, which is twice as much as your GPU.
You can try to (i) reduce the batch size, or (ii) reduce the image size. Let me know how it goes!
Hello, Thank you very much for your help! I tried scale the images, instead of (800, 1440), i used (600, 1080) and it worked. I got the following results, which are slightly different from the results in a log file i found on another github issue. Is there a chance that the reduced size is going to affect the performance of my implementation?
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.387
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=1000 ] = 0.567
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=1000 ] = 0.430
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=1000 ] = 0.279
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=1000 ] = 0.719
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=1000 ] = 0.796
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.450
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=300 ] = 0.450
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=1000 ] = 0.450
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=1000 ] = 0.350
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=1000 ] = 0.752
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=1000 ] = 0.824
01/28 11:38:15 - mmengine - INFO -
+------------+-------+------------+-------+----------+-------+
| category | AP | category | AP | category | AP |
+------------+-------+------------+-------+----------+-------+
| pedestrian | 0.411 | car | 0.495 | truck | 0.512 |
| bus | 0.411 | motorcycle | 0.428 | bicycle | 0.063 |
+------------+-------+------------+-------+----------+-------+
01/28 11:38:15 - mmengine - INFO - bbox_mAP_copypaste: 0.387 0.567 0.430 0.279 0.719 0.796
01/28 11:38:15 - mmengine - INFO - Epoch(test) [2400/2400] coco/pedestrian_precision: 0.4110 coco/car_precision: 0.4950 coco/truck_precision: 0.5120 coco/bus_precision: 0.4110 coco/motorcycle_precision: 0.4280 coco/bicycle_precision: 0.0630 coco/bbox_mAP: 0.3870 coco/bbox_mAP_50: 0.5670 coco/bbox_mAP_75: 0.4300 coco/bbox_mAP_s: 0.2790 coco/bbox_mAP_m: 0.7190 coco/bbox_mAP_l: 0.7960 data_time: 0.0017 time: 1.3705
Hi,
Thanks for sharing your results. It is very likely that the image size will affect the performance of the method. For a fair comparison, you could compare to baselines using the same image size.
Hello, Thank you for your feedback and suggestion. To ensure a fair comparison, I will indeed take into consideration using the same image size for the baselines in my evaluation.
Hello, I would like to ask, can i use an RTX 4070 12GB for evaluation and testing? When i am trying to run the mean-teacher adaptation that is provided in this repositiry i get the following error:
RuntimeError: CUDA out of memory. Tried to allocate 80.00 MiB (GPU 0; 11.72 GiB total capacity; 9.59 GiB already allocated; 47.88 MiB free; 9.81 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF.
Can i do something about it or should i use RTX 3090 GPU?
Thank you for your time!