python - Training ResNetv1 From Scratch using Tensorflow Slim -


although stated in slim model train_image_classifier.py can used train models scratch, found hard in practice. in case, trying train resnet scratch on local machine 6xk80s. used this:

dataset_dir=/nv/hmart1/ashaban6/scratch/data/imagenet_rf_record train_dir=/nv/hmart1/ashaban6/scratch/train_dir depth=50 num_clones=8  cuda_visible_devices="0,1,2,3,4,5,6,7,8" python train_image_classifier.py --train_dir=${train_dir} --dataset_name=imagenet --model_name=resnet_v1_${depth} --max_number_of_steps=100000000 --batch_size=32 --learning_rate=0.1 --learning_rate_decay_type=exponential --dataset_split_name=train --dataset_dir=${dataset_dir} --optimizer=momentum --momentum=0.9 --learning_rate_decay_factor=0.1 --num_epochs_per_decay=30 --weight_decay=0.0001 --num_readers=12 --num_clones=$num_clones

i followed same settings suggested in paper. using 8 gpus on local machine batch_size 32 effective batch size 32x8=256. learning rate set 0.1 , decayed 10 every 30 epochs. after 70k steps (70000x256/1.2e6 ~ 15 epochs), top-1 performance on validation set low ~14% while should around 50% after many iterations. used command top-1 performance:

dataset_dir=/nv/hmart1/ashaban6/scratch/data/imagenet_rf_record checkpoint_file=/nv/hmart1/ashaban6/scratch/train_dir/ depth=50  cuda_visible_devices="10" python eval_image_classifier.py --alsologtostderr --checkpoint_path=${checkpoint_file} --dataset_dir=${dataset_dir} --dataset_name=imagenet --dataset_split_name=validation --model_name=resnet_v1_${depth}

with lack of working examples hard if there bug in slim training code or problem in script. wrong in script? has trained resent scratch?


Comments

Post a Comment

Popular posts from this blog

Is there a better way to structure post methods in Class Based Views -

i'm using class based views in django , never specify url in ajax calls , omit action parameter in form because know these requests go through post method of class based view. notice view - profile page make many requests depending on user updating on page - post method become convoluted , if/else mess example: def post(self, request, *args, **kwargs): # handle user changing available date if request.post.get('availabledate') != none: self.updateavailabledate(request) return jsonresponse({'result':'success'}) if request.post.get('newprojectname') != none: creative_user = creativeuserprofile.objects.get(id = request.user.id) project = project.create_project(creative_user, request.post['newprojectname']) project.save() return jsonresponse({'projectid': project.id }) if request.post.get('projectimage') != none , request.files['file']: proj
Read more

performance - Why is XCHG reg, reg a 3 micro-op instruction on modern Intel architectures? -

i'm doing micro-optimization on performance critical part of code , came across sequence of instructions (in at&t syntax): add %rax, %rbx mov %rdx, %rax mov %rbx, %rdx i thought had use case xchg allow me shave instruction , write: add %rbx, %rax xchg %rax, %rdx however, dimay found agner fog's instruction tables , xchg 3 micro-op instruction 2 cycle latency on sandy bridge, ivy bridge, broadwell, haswell , skylake. 3 whole micro-ops , 2 cycles of latency! 3 micro-ops throws off 4-1-1-1 cadence , 2 cycle latency makes worse original in best case since last 2 instructions in original might execute in parallel. now... cpu might breaking instruction micro-ops equivalent to: mov %rax, %tmp mov %rdx, %rax mov %tmp, %rdx where tmp anonymous internal register , suppose last 2 micro-ops run in parallel latency 2 cycles. given register renaming occurs on these micro-architectures, though, doesn't make sense me done way. why wouldn't register renamer
Read more

jquery - Responsive Navbar with Sub Navbar -

good day, forgive me nav vocabulary not best. if question confusing apologize. i'm working on navigation bar needs centered on page @ 768px , above. items in nav open (or perhaps toggle) subnav lives directly below. these nav should appear centered on page. i have been able work through responsive portion of nav bar, having main items align vertically , sub nav menu of each item roll out below, , gave them indent separation. working well. however, issue not responsive side, behavior @ width greater 768px. happens toggling menu's sub navigation pushes other main items down page below sub nav. here's code: $(document).ready(function () { //stick in fixed 100% height behind navbar don't wrap $('#slide-nav.navbar-inverse').after($('<div class="inverse" id="navbar-height-col"></div>')); $('#slide-nav.navbar-default').after($('<div id="navbar-height-col"></d
Read more